1 
 

Impact of dynamical regionalization on precipitation 1 

biases and teleconnections over West Africa 2 

Iñigo Gómara1,2,3*, Elsa Mohino1, Teresa Losada1, Marta Domínguez1,2, Roberto Suárez-Moreno1,2 and Belén 3 

Rodríguez-Fonseca1,2 4 

 5 

1 Dpto. Geofísica y Meteorología, Universidad Complutense de Madrid, Madrid, Spain 6 

 7 

2 Instituto de Geociencias (IGEO), UCM, CSIC, Madrid, Spain 8 

 9 

3 CEIGRAM, Universidad Politécnica de Madrid, Madrid, Spain 10 

 11 

 12 

 13 

 14 

 15 

 16 

 17 

Submitted to 18 

Climate Dynamics 19 

11 April 2017 20 

Revised on 16 August 2017 21 

 22 

 23 

                                                           

*Correspondence to: Iñigo Gómara, Dpto. Geofísica y Meteorología, Universidad Complutense de Madrid, 

Facultad de CC. Físicas, Ciudad Universitaria s/n. 28040 Madrid, Spain. E-mail: i.gomara@ucm.es 



2 
 

Abstract 24 

West African societies are highly dependent on the West African Monsoon (WAM). Thus, a correct representation 25 

of the WAM in climate models is of paramount importance. In this article, the ability of 8 CMIP5 historical 26 

General Circulation Models (GCMs) and 4 CORDEX-Africa Regional Climate Models (RCMs) to characterize 27 

the WAM dynamics and variability is assessed for the period July-August-September 1979-2004. Simulations are 28 

compared with observations. Uncertainties in RCM performance and lateral boundary conditions are assessed 29 

individually. 30 

Results show that both GCMs and RCMs have trouble to simulate the northward migration of the Intertropical 31 

Convergence Zone in boreal summer. The greatest bias improvements are obtained after regionalization of the 32 

most inaccurate GCM simulations. To assess WAM variability, a Maximum Covariance Analysis is performed 33 

between Sea Surface Temperature and precipitation anomalies in observations, GCM and RCM simulations. The 34 

assessed variability patterns are: El Niño-Southern Oscillation (ENSO); the eastern Mediterranean (MED); and 35 

the Atlantic Equatorial Mode (EM). Evidence is given that regionalization of the ENSO-WAM teleconnection 36 

does not provide any added value. Unlike GCMs, RCMs are unable to precisely represent the ENSO impact on 37 

air subsidence over West Africa. Contrastingly, the simulation of the MED-WAM teleconnection is improved 38 

after regionalization. Humidity advection and convergence over the Sahel area are better simulated by RCMs. 39 

Finally, no robust conclusions can be determined for the EM-WAM teleconnection, which cannot be isolated for 40 

the 1979-2004 period. The novel results in this article will help to select the most appropriate RCM simulations 41 

to study WAM teleconnections. 42 

 43 

Keywords: Precipitation; West African Monsoon; Tropical Variability; Teleconnections; CMIP5; CORDEX. 44 

 45 



3 
 

1. Introduction 46 

The monsoon is the most prominent climate feature of Western Africa during boreal summer, and highly 47 

determines socio-economic development over this region (Janicot et al. 2011; Parry et al. 2007). First, because 48 

most of the annual total rainfall accumulates during a period of few months (Le Barbé and Lebel 1997; Laurent 49 

et al. 1998). Second, because West African societies are highly vulnerable to water availability, and its 50 

scarcity/excess can lead to strong agricultural losses, famine and disease spread (Dilley et al. 2005; Cook 2008; 51 

Yaka et al. 2008).  52 

The West African Monsoon (WAM) is driven by sea-land contrast of temperature and surface pressure between 53 

the Gulf of Guinea and the Sahara Desert. Therefore, during the summer months, North African continental areas 54 

are heated up more rapidly than oceanic waters (Fontaine et al. 1998; Rowell 2001). Consequently, the 55 

Intertropical Convergence Zone (ITCZ) is pushed northward and humidity is primarily advected towards the Sahel 56 

by the monsoon flux. To a lower extent, the Saharan Heat Low also plays a role in this process (SHL; Rodríguez-57 

Fonseca et al. 2011). Additional to its strong seasonal cycle, the WAM also shows remarkable variability at 58 

interannual and multi-decadal timescales, in both cases mainly due to global Sea Surface Temperature (SST) 59 

anomalies (Folland et al. 1986; Hoerling et al. 2006; Biasutti and Giannini 2006; Mohino et al. 2011a; Rodríguez-60 

Fonseca et al. 2011; 2015).  61 

At interannual timescales, three of the most prominent SST patterns influencing the WAM dynamics are: (i) El 62 

Niño-Southern Oscillation (ENSO; Janicot et al. 2001; Rowell 2001; Joly and Voldoire 2009; Mohino et al. 2011b; 63 

Diatta and Fink 2014); (ii) the eastern Mediterranean (MED; Rowell 2003; Polo et al. 2008; Fontaine et al. 2010); 64 

and (iii) the Atlantic Equatorial Mode (EM; Zebiak 1993; Rowell et al. 1995; Janicot et al. 1998; Losada et al. 65 

2010). According to the available literature, during a positive ENSO event, the warmer waters of the equatorial 66 

Pacific trigger a Kelvin wave which, over West Africa, is associated with increased air subsidence and reduced 67 

rainfall (Semazzi et al. 1988; Moron and Ward 1998; Janicot et al. 2001; Rowell 2001; Mohino et al. 2011b). The 68 

contrary holds for a negative ENSO event. This is one of several proposed mechanisms for the ENSO-WAM 69 

teleconnection (Joly and Voldoire 2009). For the Mediterranean, a warming in the eastern basin is known to 70 

increase evaporation and enhance moisture advection towards the Sahel (Polo et al. 2008; Fontaine et al. 2010; 71 

Gaetani et al. 2010). Such effect, combined with the circulation associated with the SHL, and the further 72 

penetration of the monsoonal flow over northern Africa, amplifies moisture convergence and precipitation over 73 

the Sahel (Rodríguez-Fonseca et al. 2011). Lastly, for the EM-WAM teleconnection, warmer SSTs over the 74 

equatorial Atlantic act to mitigate the pressure gradient between the Gulf of Guinea and Sahara during boreal 75 



4 
 

summer. As a consequence, when an isolated positive EM event is present, precipitation is enhanced over the Gulf 76 

of Guinea and reduced over the Sahel (Rowell et al. 1995; Janicot et al. 1998; Vizy and Cook 2001; Losada et al. 77 

2010). The opposite holds for a negative phase EM. However, during the last decades of the 20th century, both 78 

ENSO and EM patterns have tended to emerge coincidentally and in counter-phase (Rodríguez-Fonseca et al. 79 

2009). In response to both simultaneous forcings, the characteristic dipolar precipitation pattern associated with 80 

an isolated positive (negative) EM pattern has turned into a homogeneous mode of enhanced (reduced) rainfall 81 

over West Africa (Mohino et al. 2011c; Losada et al. 2012; Suárez-Moreno and Rodríguez-Fonseca 2015). 82 

Within the framework of climate research, coupled Ocean-Atmosphere General Circulation Models (GCMs) are 83 

a useful tool to characterize present/future climate dynamics and variability (IPCC AR5). However, due to the 84 

high computational costs, dynamical processes are typically resolved over a coarse horizontal grid of around 100-85 

200 km. Therefore, important dynamical features of the WAM such as Mesoscale Convective Systems (MCSs) 86 

or fine scale processes (e.g., soil-atmosphere interactions, land use) are disregarded by GCMs (Cook 2008; Steiner 87 

et al. 2009; Domínguez et al. 2010; Hourdin et al. 2010; Sylla et al. 2010). Additionally, the existence of large 88 

positive SST biases over the equatorial Atlantic in GCMs leads to a too southward representation of the Atlantic 89 

ITCZ branch during boreal summer (Richter and Xie 2008; Li and Xie 2012; Xu et al. 2014). In this context, the 90 

Coupled Model Inter-comparison Project phase 5 (CMIP5) initiative has recently coordinated efforts between 91 

international scientific communities and end users to improve present and future GCM simulations (Taylor et al. 92 

2012).  93 

The growing demand of policy makers for regionalized projections has led to the foundation of similar initiatives 94 

based on Regional Climate Models (RCMs), such as the Coordinated Regional Climate Downscaling Experiment 95 

(CORDEX; Giorgi et al. 2009; Jones et al. 2011). In this framework, CORDEX Africa prioritizes efforts on 96 

dynamical regionalization of the whole African continent (Nikulin et al. 2012). This initiative has motivated 97 

several analyses on precipitation dynamics and variability over different African Sectors: South Africa (Meque 98 

and Abiodun 2015; Favre et al. 2016), Eastern Africa (Endris et al. 2016), West Africa (Diallo et al. 2013; 99 

Gbobaniyi et al. 2014; Diasso and Abiodun 2015; Akinasola et al. 2015; Adeniyi and Dilau 2016), etc. Whereas 100 

most of these have focused on RCM simulations driven by ERA-Interim, a few less have evaluated GCM driven 101 

regionalizations. In this context, it is well known that RCMs are able to provide added value (AV) in comparison 102 

to GCMs over local areas due to resolved high resolution processes (Giorgi and Mearns 1999; Di Luca et al. 103 

2013). However, it is not so clear whether RCMs can generate AV for large-scale processes. Whereas some 104 

authors promote the use of large-scale nudging toward the lateral boundary conditions provided by GCMs (Castro 105 



5 
 

et al. 2005; Laprise et al. 2008), others dissuade against such methodology and adduce that large-scale AV can be 106 

generated by RCM internal dynamics whenever the regionalized area is sufficiently large (Mesinger et al. 2002; 107 

Veljovic et al. 2010). In any case, high resolution outputs can be meaningful only if the simulated large-scale is 108 

realistic or, at least, not worse than in the forcing GCM. Hence, both lateral boundary conditions and RCM 109 

performance are pivotal aspects for the correct representation of the large-scale circulation, and therefore 110 

teleconnections. 111 

In order to shed light upon this debate, the ability of GCMs and RCMs to represent the West African Monsoon 112 

and its most prominent teleconnection patterns is analyzed in this article. This is done through the analysis of 113 

CMIP5 historical and CORDEX-Africa simulations, accounting for uncertainties in RCM lateral boundary 114 

conditions and performance. For this purpose, a Maximum Covariance Analysis (MCA) between seasonal 115 

anomalous SSTs and precipitation is performed through the use of the novel Sea Surface Temperature based 116 

Statistical Seasonal Forecast model (S4CAST; Suárez-Moreno and Rodríguez-Fonseca 2015). 117 

The present article is organized as follows. Data and Methodology are provided in sections 2 and 3. Section 4 118 

describes the climatological aspects and most prominent teleconnections of the WAM in observations, GCMs and 119 

RCMs. The article concludes with a summary and a discussion of the main results. 120 

 121 

2. Data 122 

In this study both observations and climate model simulations are utilized.  123 

2.1 Observations 124 

Three different gridded datasets are used to analyze monthly precipitation and large-scale dynamics over West 125 

Africa: 126 

1. The NCEP/NCAR Global Precipitation Climatology Project (GPCP) version 2.1: This dataset starts in 127 

January 1979 and has a horizontal resolution of 2.5 degrees in longitude/latitude worldwide. GPCP 128 

version 2.1 is based on multi-satellite measurements and gauge observations (Huffman et al. 2009). 129 

2. The NASA Modern Era-Retrospective Analysis for Research and Application (MERRA): Based on the 130 

Goddard Earth Observing System version 5 (GEOS-5), it spans from 1979 to the present and covers the 131 

entire globe at a horizontal resolution of 0.5/0.67 degrees in longitude/latitude (Rienecker et al. 2011). 132 

3. The European Centre for Medium-Range Weather Forecasts Interim Re-analysis (ERA-Interim): It 133 

prolongs from 1979 to the present and features a worldwide horizontal resolution of 0.75 degrees in 134 

longitude/latitude along 60 vertical levels (Dee et al. 2011). 135 



6 
 

Additionally, sea surface temperature and sea ice data from the Met Office Hadley Centre (HadISST) are utilized. 136 

This dataset spans from 1870 to the present on monthly basis and has a horizontal resolution of one degree in 137 

longitude/latitude. HadISST is based on the Met Office Marine Data Bank and the Global Telecommunications 138 

System, among other data sources (Rayner et al. 2003). 139 

2.2 Model Simulations 140 

Historical simulations of 8 different GCMs from the CMIP5 initiative are analyzed. The simulations cover the last 141 

century and a half (1861-2005) and are mainly forced by observations of atmospheric gas composition (including 142 

greenhouse gases) and land cover (Taylor et al. 2012). The model names and modeling institutions are provided 143 

in Table 1. 144 

Additionally, 11 RCM simulations from the CORDEX-Africa experiment are studied. All models cover the period 145 

1951-2005 over the African continent at a horizontal resolution of 0.44 degrees in both longitude and latitude (cf. 146 

Fig. S1a; Giorgi et al. 2009; Nikulin et al. 2012). The nested RCM simulations are separated into two different 147 

experiment blocks depending on the GCM-RCM configuration. First, a set of 8 different GCM historical 148 

simulations driving the same regional model (SMHI-RCA4) is considered. This is done to assess uncertainties in 149 

RCM boundary lateral conditions. Second, the same GCM model (MPI-ESM-LR) is set as boundary condition 150 

for 4 different regional models (CCLM4-8, CSC-REMO, SMHI-RCA4 and UQAM-CRCM5). Here, the 151 

performance of different nested RCMs is analyzed based on the same external forcing. For clarity, the GCM-152 

RCM combinations considered in this article are specified in Table 1. Hereafter we will refer as GCMs, GCMs-153 

RCA4, MPI and MPI-RCMs to the GCM-RCM block experiments described in this section. 154 

 155 

3. Methods 156 

In order to analyze the mean state and teleconnections of the WAM, the peak precipitation period (July-August-157 

September; JAS) over the West African area [WA; 20ºW-30ºE, 0º-20ºN] is selected. The choice of area follows 158 

Sultan and Janicot (2000) and Janicot et al. (2011). Additionally, two smaller regions representative of West 159 

Africa North [WA-N; 10ºW-10ºE, 9º-15ºN] and West Africa South [WA-S; 10ºW-10ºE, 5º-9ºN] are considered 160 

(cf. boxes in Fig. 1a). In these regions, human socio-economic activity is highly linked to precipitation, which is 161 

homogeneous all over their territory (cf. CORDEX protocol; Nikulin et al. 2012; Laprise et al. 2013; Gbobaniyi 162 

et al. 2014; Dosio et al. 2015). The period of study covers 26 consecutive JAS seasons, from 1979 to 2004, which 163 

is the common time interval for model and observational data. Year 2005 is not included in the analysis due to the 164 

unavailability of data for some particular models and variables. 165 



7 
 

3.1 Interpolation of the fields 166 

A bilinear interpolation is applied to all fields of this study (e.g., precipitation, Sea Surface Temperature, wind 167 

divergence, etc.), which are regridded to a 2.8º x 2.8º horizontal resolution. The grid corresponds to the GCM of 168 

coarser resolution (CanESM2). This is done to allow comparison between fields from different sources (GCMs, 169 

RCMs and observations) and to avoid issues derived from increasing data resolution artificially (Wilks 2006). 170 

Thus, the remapping of RCM fields returns clearer large-scale spatial patterns at the cost of obscuring fine-scale 171 

processes. Nevertheless, the main focus of this paper is more on the large-scale atmospheric circulation and less 172 

on local-scale features. To keep consistency among calculations, all SST maps are also downgraded to the 173 

resolution of CanESM2. 174 

3.2 Calculation of Ensembles 175 

Ensemble means are provided along this article for the GCM, GCM-RCA4 and MPI-RCM block experiments. In 176 

each experiment block the output from each model simulation (1 realization per model) is averaged over the rest. 177 

Robustness of ensembles is calculated from the number of members of each class depicting the same sign of a 178 

given variable (e.g., precipitation anomaly, regression map, etc.). 179 

3.3 Empirical Orthogonal Functions and Maximum Covariance Analysis 180 

Empirical Orthogonal Functions (EOF) analysis is a powerful discriminant statistical method that determines the 181 

most prominent modes of variability of a given anomalous field (Lorenz 1956). In this article, EOF analysis is 182 

performed on SST and precipitation anomaly fields. To evaluate the co-variability of West African rainfall with 183 

Tropical Pacific, Mediterranean and Tropical Atlantic SST anomalies, a Maximum Covariance Analysis (MCA) 184 

is also utilized (Bretherton et al. 1992; Widmann 2005). The latter is often considered as a generalization of EOFs. 185 

MCA is performed between a predictor Y field (SST anomalies) and a standardized predictand Z field 186 

(precipitation anomalies) over specific regions and time periods. The method applies a Singular Value 187 

Decomposition to the non-square covariance matrix of both fields, and performs linear combinations between the 188 

time series of Y and Z (a.k.a. expansion coefficients U and V, respectively) to maximize the former (Cherry 1997). 189 

Once the covariance matrix is diagonalized, the method retrieves the singular vectors R and Q, which are the 190 

leading co-variability modes between Y and Z over the regions and period selected. The squared covariance 191 

fraction (scf) of these modes is provided in %, with a confidence interval (90%) calculated through a Monte Carlo 192 

test of 1000 random iterations. The linear trend is removed from the SST predictor/predictand fields to mitigate 193 

the potential signal of anthropogenic climate change in the MCA analysis (IPCC AR5). No further filtering is 194 

applied to the fields, as the aim is to focus on interannual variability, and potential lower-frequency oscillations 195 



8 
 

(e.g., decadal, multi-decadal) are hard to detect over a period of just 26 years. For the same reason, it is also 196 

assumed that teleconnections are stationary during the study period, despite it does not seem to be the case when 197 

longer intervals are considered over specific areas: e.g., West Africa (Mohino et al. 2011c), the Tropical Atlantic 198 

(Rodríguez-Fonseca et al. 2009; 2016), the extra-tropical North Atlantic (López-Parages and Rodríguez-Fonseca 199 

2012; Raible et al. 2014; Gómara et al. 2016), etc. 200 

For the calculation of the MCA, the Sea Surface Temperature based Statistical Seasonal Forecast model (S4CAST; 201 

v.2.0), recently developed by Suárez-Moreno and Rodríguez-Fonseca (2015), is used. Although the S4CAST 202 

model was mainly conceived to work in ‘forecast’ mode (predictor field leading in time), its ‘synchronous’ mode 203 

is selected (lead time 0), as the objective is to accurately characterize concurrent climate anomalies in 204 

observations, GCMs and RCMs. 205 

3.4 RCM Added Value 206 

RCM Added Value (AV) is calculated throughout this article for different variables. This measure provides 207 

information on how well RCMs are able to reproduce a given field in comparison with GCMs, considering gridded 208 

observations as basis. The AV is computed as follows (Di Luca et al. 2013; Meque and Abiodun 2015): 209 

 210 

𝐴𝑉 = (𝑋𝐺𝐶𝑀 − 𝑋𝑂𝐵𝑆)2 −  (𝑋𝑅𝐶𝑀 − 𝑋𝑂𝐵𝑆)2,                                                    (1) 211 

 212 

Where XGCM, XRCM and XOBS are the corresponding fields for the GCM, RCM and observations, respectively.  213 

Thus, if AV is positive (negative), the RCM improves (deteriorates) GCM simulations, taking observations as 214 

reference. 215 

 216 

4. Results 217 

4.1 Seasonal Precipitation Biases 218 

In this section, seasonal precipitation biases of GCM, GCM-RCA4, and MPI-RCM simulations are analyzed for 219 

the period JAS 1979-2004. The seasonal mean precipitation of the observational GPCP dataset, which is chosen 220 

as reference, is provided in Fig. 1a. The characteristic zonal rainfall belt extending from West Africa to 221 

Chad/Sudan is present in the figure, with two regional maxima situated over the Guinea-Conakry/Sierra Leone 222 

coast and the Cameroon Highlands (Cook and Vizy 2006). The standard deviation field appears overlaid in Fig. 223 

1a. The spatial overlapping between the mean and standard deviation indicates that the strongest variability is 224 

enclosed over the areas of highest precipitation. These patterns are similar to that obtained from MERRA (Fig. 225 



9 
 

S1b). Nevertheless, it is worth to mention that GPCP tends to produce higher precipitation rates over the Central 226 

African Republic, Chad and surrounding area (Fig. S1c). This inconsistency has been mentioned in previous 227 

studies and appears to be caused by different gauge station availability over the area in GPCP and MERRA 228 

datasets (Huffman et al. 2009; Yin and Gruber 2010; Nikulin et al. 2012). 229 

Next, the ensemble precipitation bias of GCM historical runs is shown (Fig. 1b). As it can be observed, GCMs 230 

have difficulties to accurately represent the northward migration of the ITCZ in boreal summer and its associated 231 

rain belt. As a consequence, rainfall amounts and variability are notably overestimated south of the Gulf of Guinea 232 

coast. The contrary is observed over the westernmost Sahel, especially over Senegal and Gambia. As mentioned 233 

in the introduction, this is caused by the existence of warm SST biases over the tropical Atlantic in GCMs (Fig. 234 

S1d), which act to mitigate the sea-land pressure gradient between the Gulf of Guinea and the Sahara Desert 235 

(Richter et al. 2012). For completeness, the individual seasonal precipitation biases of GCM historical runs are 236 

presented in Figs. S2a-h. In this context, HadGEM2-ES (Fig. S2e) and NorESM1-M (Fig. S2h) are clear examples 237 

of a too southward representation of the ITCZ. Oppositely, MPI-ESM-LR (a.k.a. MPI) provides one of the best 238 

estimates of the WAM over West Africa (Fig. S2g). 239 

For comparison, the ensemble precipitation bias of GCM-RCA4 simulations is provided in Fig. 1c. Rainfall is 240 

again overestimated over the Gulf of Guinea and underestimated over Senegal. However, the zonal band of 241 

inflated precipitation appears stronger and narrower (compare Figs. 1b and 1c). In addition, precipitation is 242 

evidently underestimated over the Congo region (Fig. 1c), an aspect not observed in GCMs (Fig. 1b). Such error 243 

is present systematically in all GCM-RCA4 individual simulations (Figs. S2i-p), and may be inherent to the 244 

SMHI-RCA performance. Although for a slightly different time period, model version and lateral boundary 245 

conditions (ERA-Interim), the same dry bias over the Congo area was observed between SMHI-RCA35 and GPCP 246 

in Nikulin et al. (2012; their Fig. 4). 247 

Next, the ensemble of MPI-RCMs seasonal precipitation bias and its individual components are provided in Figs. 248 

1d and S2q-t, respectively. The selection of MPI as boundary condition for the RCM runs is based on the relative 249 

good performance of this model representing the WAM (Fig. S2g). Consistent with the previous GCM-RCA4 250 

simulations, most of RCMs tend to exaggerate precipitation south of the Guinean coast (Fig. 1d). Regarding the 251 

dry bias over the Congo region, it is again evident in the MPI-RCM ensemble, although of weaker intensity 252 

(compare Figs. 1c and 1d). Attending to the individual runs (Figs. S2q-t), this appears to be a common issue in 253 

RCMs, and not only inherent to SMHI-RCA4.  254 



10 
 

Finally, the RCM added value of seasonal precipitation is provided in Figs. 1e-f (ensembles) and Fig. S3 255 

(individual runs). For the GCMs-RCA4 ensemble, the regional model is able to minimize mean precipitation 256 

biases inland over West Africa and the Tropical South Atlantic (Fig. 1e). Average negative AVs are present south 257 

of the Guinean coast and over the Congo area. Individual AVs of GCM-RCA4 runs are available in Figs. S3a-h. 258 

Both HadGEM2-ES/RCA4 (Fig. S3e) and MIROC5/RCA4 (Fig. S3f) runs are examples of good performance 259 

over West Africa and the Gulf of Guinea. The opposite case is GFDL-ESM2M/RCA4 (Fig. S3d), which 260 

remarkably fails to represent precipitation south of the Guinean coast and is the main contributor of the ensemble 261 

negative AVs over this area in Fig. 1e. Regarding MPI-RCM simulations (Fig. 1f), the ensemble AV shows 262 

negative values all over West Africa’s coast and mainland. The scarce positive values are present over the 263 

equatorial and Tropical South Atlantic. Attending to the equation of Added Value exclusively (1), one possibility 264 

for these results is that the MPI simulation already provides quite accurate values of the WAM rainfall amounts 265 

(Fig. S4g). Consequently, little room is left for RCM improvement (Meque and Abiodun 2015). Nevertheless, the 266 

key factor behind GCM/RCM precipitation biases is the role of physics in the simulations considered. Due to 267 

resolution limitations, convective rainfall is parameterized in both global and CORDEX-Africa simulations 268 

analyzed, but the convection scheme may not be similar between the driving GCM and the regional model in 269 

some cases (Nikulin et al. 2012). As rainfall variability is highly sensitive to the physical package utilized for 270 

convection (and other parameterized processes; Flaounas et al. 2011), a significant/systematic improvement of 271 

precipitation biases in all RCM simulations is hard to be accomplished. Anyhow, these conclusions may be 272 

different in convection-permitting or cloud-resolving higher resolution simulations, where a significant 273 

improvement should be expected after regionalization (Randall et al. 2003; Prein et al. 2015). Lastly, the 274 

individual AVs of MPI-RCM simulations are available in Figs. S3i-l. 275 

So far, precipitation biases have been spatially characterized over West Africa. Subsequently, average 276 

precipitation biases over West Africa North [WA-N; 10ºW-10ºE, 9º-15ºN], West Africa South [WA-S; 10ºW-277 

10ºE, 5º-9ºN] and the whole West Africa domain [WA; 20ºW-30ºE, 0ºN-20ºN; cf. Sections 4.2 to 4.4] are 278 

quantified in Fig. 2. The mean values corresponding to the MERRA dataset in both regions are included as well 279 

in Fig. 2. In general, GCM-RCA4 simulations tend to reduce mean seasonal precipitation biases of GCMs over 280 

WA-N, WA-S and WA. These results can be inferred from the ‘Ensemble GCMs’ bars in Figs. 2a,c,e and are 281 

valid for both GPCP and MERRA. The contrary holds for MPI and MPI-RCMs over WA-S and WA, where 282 

precipitation biases are increased after regionalization. This can be extracted from the ‘Ensemble MPI’ bars of the 283 

same figures (GPCP and MERRA). For WA-N, the mean seasonal precipitation biases are fairly weak in both 284 



11 
 

MPI and MPI-RCM simulations. In general terms, results for the whole WA are similar to those from WA-S, but 285 

with lower values (Figs. 2c,e). 286 

Regarding standard deviation, most of GCM-RCA4 members tend to underestimate the year-to-year variability in 287 

seasonal precipitation over WA-N (Fig. 2b), while the reverse holds for WA-S (Fig. 2d). Results are disparate for 288 

the whole WA domain (Fig. 2f). Regarding the MPI-RCMs, most of members overestimate the year-to-year 289 

variability in seasonal precipitation in WA-S and WA (Figs. 2d,f). Among them, MPI-CCLM4 and MPI-REMO 290 

biases largely exceed the observational uncertainty in all regions, suggesting that these two models control the 291 

precipitation variability of the 'MPI Ensembles'. Due to the large differences in standard deviation observed 292 

between GPCP and MERRA over all regions (horizontal red lines; Fig. 2 - right column), no further conclusions 293 

can be here obtained. 294 

So far, climatological GCM and RCM simulations of the WAM have been compared with observations. In the 295 

following, a similar analysis is carried out but focusing on the representation of the most prominent interannual 296 

variability SST patterns influencing the WAM (cf. Section 1). 297 

4.2 Influence of El Niño-Southern Oscillation on the West African Monsoon in observations, GCMs and RCMs 298 

In this section the impact of ENSO on interannual WAM variability is assessed in observations, GCMs and RCMs. 299 

For this purpose, a MCA between seasonal SST anomalies over the Equatorial Pacific [110ºE-80ºW, 20ºS-20ºN; 300 

predictor field] and simultaneous precipitation anomalies over West Africa [20ºW-30ºE, 0º-20ºN; predictand 301 

field] is performed. The linear trend is removed from the predictor/predictand fields to mitigate the potential 302 

influence of anthropogenic climate change in the results. 303 

(A) Observations 304 

In Figs. 3a-b the SST homogeneous and precipitation heterogeneous maps are provided for the HadISST-GPCP 305 

gridded observational datasets. The leading SST mode depicts a well-defined and robust positive phase ENSO 306 

pattern (shadings/stippling in Fig. 3a), which accounts for 41% of total squared covariance fraction (scf) and 58% 307 

of explained SST variance over the Equatorial Pacific (EOF1; cf. Fig. S4a and Table S1). The pattern in Fig. 3a 308 

is accompanied in the Equatorial Atlantic by much weaker negative SST anomalies (Rodríguez-Fonseca et al. 309 

2009; Losada et al. 2012). The associated heterogeneous map reveals a large-scale and statistically significant 310 

pattern of decreased precipitation that extends along the corridor 0º-20ºN over Western and Central Africa (Fig. 311 

3b). This pattern corresponds to EOF1 of seasonal precipitation over West Africa (precipitation exp. var. 30%; cf. 312 

Fig. S4d and Table S2). By construction, the co-variability modes are also valid if a minus sign is applied to both 313 

maps (i.e., negative ENSO phase & increased precipitation over Western/Central Africa). In order to characterize 314 



12 
 

the underlying dynamics of the ENSO-WAM teleconnection, the projection of expansion coefficient U on two 315 

different dynamical fields from ERA-Interim is given in Fig. 3c. The selected fields are anomalous seasonal wind 316 

divergence (shadings) and velocity potential (contours). To characterize the fields at both lower and upper levels, 317 

anomalies from pressure level 850 hPa are subtracted to those from 200 hPa (hereafter DIV200/850 and 318 

KHI200/850). Therefore, positive regression anomalies of DIV200/850 will generally be associated with 319 

enhanced air divergence at 200 hPa and convergence at 850 (i.e., increased air uplift). The contrary holds for 320 

negative DIV200/850 anomalies (enhanced air subsidence). For velocity potential, positive (negative) anomalies 321 

are associated with intensified downward (upward) air movement.  322 

Attending to Fig. 3c, a positive ENSO event is linked to enhanced air subsidence over the Gulf of Guinea and the 323 

West Africa corridor 0º-20ºN. Please note that the color bar in Fig. 3c is reversed to improve visual comparison 324 

with Fig. 3b. To permit a wider perspective, the regression map of KHI200/850 is provided globally in Fig. 3a 325 

(contours). As expected, the outcome from Figs. 3a-c is in good agreement with the so-called “ENSO-WAM 326 

Kelvin wave teleconnection mechanism”, already described in previous studies (Folland et al. 1986; Palmer 1986; 327 

Janicot et al. 1996; 2001; Rowell 2001; Joly and Voldoire 2009; Mohino et al. 2011b). Finally, the negative SST 328 

anomalies present over the equatorial Atlantic might, at some point, have an imprint on West Africa precipitation. 329 

However, the ENSO related SST anomalies are of much stronger intensity and seem to dominate the 330 

teleconnection (Fig. 3). This is consistent with Losada et al. (2012). 331 

(B) GCMs & GCMs-RCA4 332 

As a next step, the ENSO-WAM teleconnection is assessed on the historical GCM runs (Joly et al. 2007). For 333 

simplicity, ensemble GCM maps are discussed first and provided in the main manuscript (Fig. 4, left column). 334 

Results for individual simulations are available in the supplementary material (Fig. S5). 335 

The GCM ensemble homogeneous SST and heterogeneous precipitation maps are provided in Figs. 4a,c. On the 336 

one hand, a robust positive ENSO pattern (32% scf), very similar to that obtained from observations, is present in 337 

Fig. 4a (stippling indicates 7/8 models with same sign on regression). A concomitant negative phase EM over the 338 

eastern tropical Atlantic is also present. On the other hand, two areas of decreased precipitation can be seen in the 339 

ensemble heterogeneous map (Fig. 4c). The first is situated over the Equatorial Atlantic. The second stretches 340 

along 15ºN over Western Africa. The projection of expansion coefficient U on anomalous DIV200/850 and 341 

KHI200/850 fields is provided for the GCM ensemble (Fig. 4e). The spatial overlapping between precipitation 342 

and DIV200/850 anomalies in Figs. 4c,e is very clear and corroborates the strong relation between both fields. 343 

Therefore, GCMs are also able to reproduce the ENSO-WAM teleconnection mechanism detected in observations. 344 



13 
 

However, the strongest precipitation/subsidence anomalies are shifted to the south in GCMs (compare Figs. 3b,c 345 

and 4c,e). That is consistent with a too southward representation of the ITCZ (Fig. 1b). In addition, the enhanced 346 

subsidence and decreased precipitation anomalies over the Sahel are narrower and of weaker intensity in GCMs. 347 

Regarding the individual GCM simulations (Fig. S5), in all of them ENSO represents the leading mode, with 348 

squared covariance fractions ranging from 25% (CNRM-CM5) to 45% (MIROC5). For the individual 349 

precipitation and subsidence maps, a certain degree of dissimilarity among models is found. For instance, CNRM-350 

CM5 is associated with increased precipitation and air ascent over the Guinean coast during positive ENSO events 351 

(Figs. S5d-f), whereas the anomalies are of opposite sign in HadGEM2-ES and MIROC5 under a similar external 352 

forcing (Figs. S5m-r). 353 

After having characterized the ENSO-WAM teleconnection in observations (Fig. 3) and GCMs (Fig. 4-left 354 

column), the same analysis follows for the ensemble of GCM-RCA4 simulations (Fig. 4-right column). On the 355 

one hand, the SST ensemble homogeneous map from GCMs-RCA4 is exceptionally similar to that found from 356 

GCMs (cf. Figs. 4a and 4b). That is expected as the MCAs for the GCM-RCA4 simulations are based on the same 357 

GCM predictor data (only precipitation is regionalized over Africa) and the dominance of ENSO variability over 358 

the equatorial Pacific is paramount (Trenberth 1997). On the other hand, the precipitation ensemble of GCM-359 

RCA4 heterogeneous maps is not robust in space (Fig. 4d), and only depicts multi-model agreement near the 360 

equator. Attending to the regression maps of DIV200/850 and KHI200/850, the outcome is very similar (Fig. 4f). 361 

The ENSO impact on air subsidence only remains robust near the equator and the signal over the Sahel vanishes. 362 

In this line, a strong multi-model spread is found in individual precipitation and subsidence GCM-RCA4 363 

regression maps (Fig. S6). For instance, results from CanESM2/RCA4 (Figs. S6b,c) and GFDL/RCA4 (Fig. S6k,l) 364 

are very different over the Sahel. The squared covariance fraction of the leading MCA modes from GCM-RCA4 365 

simulations also returns more spread in their values (21 to 54%). 366 

(C) MPI & MPI-RCMs 367 

Next, the ENSO-WAM teleconnection is assessed for MPI and MPI-RCM simulations. The analysis is the same 368 

as provided in (B), with the exception that MPI results (Fig. 5, left column) are specific to a single model 369 

simulation, whereas MPI-RCM ensembles are composed of 4 members (Fig. 5, right column). 370 

Attending to MPI performance, it relates an ENSO type of SST anomaly (Fig. 5a; scf 33%) with decreased 371 

precipitation (Fig. 5c) and increased air subsidence (Fig. 5e; cf. DIV200/850 in colors) over the Gulf of Guinea. 372 

However, the simulated ENSO impact over the Sahel (15ºN) is very weak, especially near the Greenwich meridian 373 

(Figs. 5c,e). 374 



14 
 

Regarding MPI-RCM results (Fig. 5, right column), the ensemble SST regression map (Fig. 5b; scf 31%) is almost 375 

identical to that obtained from MPI (Fig. 5a). This is because all simulations use the same SST forcing from MPI 376 

(cf. individual SST maps in Fig. S7). For the precipitation and air subsidence ensemble maps (Figs. 5d,f), results 377 

are very similar to those from the MPI simulation. In particular, the ENSO impact on Sahelian precipitation is 378 

even weaker in the MPI-RCMs ensemble (compare Figs. 5c and 5d). In this line, MPI-RCM individual simulations 379 

return rather different precipitation/subsidence anomalies over Western Africa (Fig. S7), as are the cases of 380 

MPI/CCLM4-8 (Figs. S7a-c) and MPI/CSC-REMO (Figs. S7d-f). 381 

(D) Added Value results 382 

As a summary of sections (A)-(C), the ensemble RCM added value of heterogeneous precipitation maps is 383 

calculated based on Eq. (1). For this purpose, the average AV from individual simulations (shadings) and the 384 

number of models in which this variable is positive at a given grid-point (contours) are shown in Fig. 6. In this 385 

case, positive RCM AVs reveal regions where the representation of the ENSO-WAM teleconnection is improved 386 

compared with GCMs (considering observations as basis).  387 

The results of GCM-RCA4 vs. GCM simulations are provided in Fig. 6a. AVs are mainly negative over the box 388 

of study [20ºW-30ºE, 0º-20ºN], and the scarce positive values appear constrained over the Equator. Very similar 389 

results are obtained for MPI-RCMs vs. MPI (Fig. 6b). Therefore, RCMs seem to deteriorate the representation of 390 

the ENSO-WAM teleconnection provided by GCMs (see also individual AV members in Fig. S8). In this context, 391 

GCMs are able to capture, although with weaker intensity, the large-scale ENSO impact on air subsidence over 392 

the Sahel (Figs. 3-4). However, RCMs are not able to improve the signal present in the lateral boundary conditions 393 

over their domain (Figs. 4-5). According to these results, the considered CORDEX-Africa RCMs are less skillful 394 

than GCMs to represent the ENSO-WAM teleconnection. At this point, it must be reminded that this study does 395 

not intend to clarify the technical reasons why RCMs improve/deteriorate GCM simulations. The main focus is 396 

to evaluate how well interannual variability modes are simulated by GCMs/RCMs and under which circumstances 397 

these simulations could be used in future research. 398 

4.3 Influence of the Mediterranean on the West African Monsoon in observations, GCMs and RCMs 399 

This section focuses on the evaluation of the MED-WAM teleconnection in observations, GCMs and RCMs. JAS 400 

seasonal SST anomalies over the whole Mediterranean domain [0º-40ºE, 30ºN-45ºN] are considered as predictor. 401 

Precipitation anomalies over the West African domain [20ºW-30ºE, 0º-20ºN] are selected as predictand. 402 

(A) Observations 403 



15 
 

Fig. 7 depicts the MCA results between HadISST and GPCP observational datasets. The leading co-variability 404 

mode explains 40% of squared covariance fraction. The homogeneous SST map shows a robust SST warming in 405 

the Mediterranean (Fig. 7a). This pattern corresponds to EOF1 of SST anomalies over the same area and explains 406 

52% of variability (Fig. S4b and Table S1). The heterogeneous rainfall map exhibits widespread, significant 407 

positive anomalies over the Sahel (Fig. 7b). Expansion coefficient V of this pattern is significantly linked to EOF1 408 

(exp. var. 30%; Fig. S4d) and EOF2 (exp. var. 25%; Fig. S4e) of precipitation anomalies over West Africa (Table 409 

S2). By linearity of the method, the opposite MCA patterns can also be considered. The dynamics of the WAM-410 

MED teleconnection are characterized in Fig. 7c, where expansion coefficient U is projected on specific humidity 411 

at 850 hPa (SHUM850, shadings), moisture flux at 850 hPa (MF850, arrows), and sea level pressure (SLP, 412 

contours) anomalies. Specifically, a warming in the Mediterranean is associated with local increased evaporation 413 

and low-level moisture, which is advected to the south by the low pressure anomalies situated over the eastern 414 

Sahara (Fig. 7c). Concurrently, the anomalous large-scale configuration strengthens the SLP gradient between the 415 

Gulf of Guinea and Sahara, thus intensifying the southwesterly monsoonal flow. In consequence, moisture supply 416 

and rainfall are increased over the Sahel area (Figs. 7b-c). These results are consistent with previous studies on 417 

this topic (Rowell 2003; Jung et al. 2006; Fontaine et al. 2010; 2011; Gaetani el al. 2010). 418 

(B) GCMs & GCMs-RCA4 419 

The reproduction of the leading SST mode by the ensemble of individual GCMs is presented in Fig. 8a. A robust 420 

warming over the Mediterranean is found (32% scf). A positive warming over the Equatorial Atlantic is also 421 

visible in the figure. The latter could be related to the ensemble rainfall positive anomalies present over the Gulf 422 

of Guinea region in Fig. 8c. In GCMs, the center of the associated low-pressure anomalies is shifted northward 423 

(over the Mediterranean Sea) compared to observations (Figs. 8e and 7c). As a consequence, the anomalous large-424 

scale configuration no longer promotes moisture inflow over the Sahel from the Mediterranean (Fig. 8e). 425 

Additionally, the anomalous southwesterly monsoonal flow is clearly underestimated in GCM simulations, 426 

probably due to the appearance of EM+ SST anomalies in the ensemble homogeneous map (Fig. 8a). As a result, 427 

humidity supply and precipitation are notably reduced over the Sahel area compared to observations (cf. Figs. 428 

7b,c and Figs. 8c,e). On individual GCM simulations (Fig. S9), SST maps generally capture a warming in the 429 

Mediterranean. Contrastingly, the positive signal over the tropical Atlantic is not reproduced by all models. An 430 

exception is the NorESM1-M model, which provides very similar results to that found in observations (Figs. S9v-431 

x). In this model, the prominent low pressure anomalies over northern Africa enhance low-level moisture 432 

advection from the western Mediterranean and Gulf of Guinea towards the Sahel (Fig. S9x). In this line, a correct 433 



16 
 

representation of the Saharan heat low seems essential to characterize the EM-WAM teleconnection (Lavaysse et 434 

al. 2010ab; Evan et al. 2015). 435 

Results for GCM-RCA4 simulations are given in Fig. 8 (right column). On the one hand, the ensemble SST 436 

homogeneous map reveals again MED+ and EM+ concurrent anomalies (Fig. 8b). On the other hand, the 437 

associated precipitation map returns widespread rainfall anomalies over the Sahel area, additional to those over 438 

the Gulf of Guinea (Fig. 8d). In this case, southwesterly moisture advection over the Sahel is better simulated than 439 

in GCMs, consistent with a strengthened Saharan/Mediterranean heat low (compare Figs. 8e-f). However, the 440 

position of the low is again improperly simulated in GCMs-RCA4, and the Mediterranean branch of moisture 441 

advection over the Sahel is absent. Thus, precipitation and SHUM850 anomalies are better simulated over the 442 

eastern Sahel, where the monsoonal moisture branch appears to dominate (Figs. 8d,f). For completeness, GCM-443 

RCA4 individual regression maps are provided in Fig. S10. The models that better capture the MED-WAM 444 

teleconnection are CNRM-CM5/RCA4 (Figs. S10d-f) and NorESM1-M/RCA4 (Figs. S10v-x). In both cases, a 445 

robust SST warming is found over the Mediterranean, while anomalies are inexistent over the Equatorial Atlantic. 446 

Moreover, in these two models the associated SLP pattern shows a strong latitudinal gradient over the Sahel, a 447 

factor that seems crucial to foster inflow humidity flux from the Gulf of Guinea. 448 

(C) MPI & MPI-RCMs 449 

The results for MPI and MPI-RCM simulations are collected in Fig. 9. The analysis is conducted as in section 450 

(B). In both experiments, ensemble homogeneous maps depict a warming of Mediterranean SSTs (Figs. 9a-b). 451 

Compared to GCMs, the concurrent positive anomalies over the equatorial Atlantic are much weaker and confined 452 

near the coast (cf. Figs. 8a-b and 9a-b). Attending to the ensemble precipitation regression maps, regionalized 453 

simulations are again able to provide a better estimate of precipitation anomalies over the Sahel (Figs. 9c-d). 454 

Whereas in the MPI simulation a precipitation dipole is found between the Guinean coast and Sahara (Fig. 9c), in 455 

the MPI-RCM ensemble widespread positive anomalies are present over the Sahel (Fig. 9d). In both cases, the 456 

limited positive precipitation anomalies over the eastern Equatorial Atlantic could be related to the SST warming 457 

over this area. The dynamics of the MED-WAM teleconnection are analyzed in Figs. 9e-f. On MPI (Fig. 9e), the 458 

anomalous large-scale SLP configuration seems to promote shallow moisture transport from the Mediterranean 459 

towards the Sahel. However, moisture advection from the Gulf of Guinea is completely blocked. As a 460 

consequence, neither low-level humidity nor precipitation appear enhanced over the Sahel. On MPI-RCMs (Fig. 461 

9f), the ensemble regression map reveals that moisture advection over the Sahel takes places throughout the 462 

northern (Mediterranean) and southern (monsoonal) branches. As a result, Sahelian low-level humidity is 463 



17 
 

significantly enhanced and precipitation is promoted. In this case, the deeper/broader negative SLP anomalies 464 

simulated by RCMs over Northern Africa might contribute to strengthen the monsoonal flow (compare Figs. 9e 465 

and 9f). According to these results, the southern branch of humidity advection seems to play a much more 466 

determinant role than the northern one on the MED-WAM teleconnection. 467 

Regarding individual MPI-RCM simulations (Fig. S11), the warming of the Mediterranean is well captured by 468 

the 4 members of the ensemble. Some differences appear in other regions though. For instance, compare 469 

MPI/CCLM4 and MPI/CRCM5 runs over the sub-tropical North Atlantic (Figs. S11a and S11j). The less realistic 470 

simulation of the MED-WAM teleconnection is given by MPI/REMO (Figs. S11d-f), which completely fails to 471 

simulate the zonal SLP gradient over West Africa.  472 

 (D) Added Value results 473 

The use of nested RCMs in reproducing the MED-WAM teleconnection is weighted by the ensemble added values 474 

of heterogeneous rainfall maps (Fig. 10). As expected, robust positive AVs are provided by GCM-RCA4 and 475 

MPI-RCM ensembles over the central and eastern Sahel. Over these areas, the Mediterranean influence on 476 

precipitation is known to be higher (cf.  Rowell 2003; Fontaine et al. 2010; Gaetani el al. 2010). For completeness, 477 

AVs for individual members can be examined in Fig. S12. Based on these results, RCMs can be considered as a 478 

useful tool to substantially improve GCM simulations of the MED-WAM teleconnection.  479 

4.4 Influence of the Atlantic Equatorial Mode on the West African Monsoon in observations, GCMs and RCMs 480 

In this section the Atlantic EM influence on WAM precipitation is analyzed. With this aim, seasonal SST 481 

anomalies over the equatorial Atlantic [60ºW-20ºE, 20ºS-5ºN] are selected as predictor field. For the predictand, 482 

the same area [20ºW-30ºE, 0º-20ºN] of standardized concurrent precipitation anomalies is chosen. 483 

 (A) Observations 484 

The MCA results for the HadISST-GPCP observational datasets are shown in Fig. 11. The SST homogeneous 485 

map reveals that the Atlantic EM (positive phase) is the leading co-variability mode, explaining 41% of squared 486 

covariance fraction (Fig. 11a). This pattern corresponds to EOF1 of SST variability over the Equatorial Atlantic 487 

(SST exp. var. 64%; Fig. S4c and Table S1). Together with a positive EM, a negative ENSO pattern is also visible 488 

in Fig. 11a over the western equatorial Pacific (Rodríguez-Fonseca et al. 2009). Unlike the SST forcing in Section 489 

4.2 (Fig. 3a), the SST anomalies over the equatorial Atlantic and Pacific are now of similar amplitude.  490 

The associated precipitation regression map is given in Fig. 11b, and depicts a homogeneous mode of positive 491 

anomalies extending from the Gulf of Guinea towards the Western Sahel (Losada et al. 2012). Expansion 492 

coefficient V of this pattern is significantly linked to EOF1 (exp. var. 30%; Fig. S4d) and EOF2 (exp. var. 25%; 493 



18 
 

Fig. S4e) of precipitation anomalies over West Africa (cf. Table S2). To analyze the underlying dynamics of this 494 

mode, seasonal anomalies of DIV200/850 (colors), MF850 (arrows) and SHUM850 (contours) are regressed on 495 

expansion coefficient U (Fig. 11c). As it can be observed, the warmer waters of the equatorial Atlantic (Fig. 11a) 496 

increase low-level specific humidity over this area (contours in Fig. 11c). As the pressure gradient between the 497 

Gulf of Guinea and Sahara is mitigated due to the Atlantic equatorial SST warming, the northward advection of 498 

moisture is restricted over the Guinean coast and surrounding area (arrows in the same figure; Janicot 1992; 499 

Fontaine and Janicot 1996). As a consequence, both DIV200/850 (Fig. 11c - colors) and precipitation (Fig. 11b) 500 

appear enhanced over this region. In this line, the presence of significant concurrent negative ENSO anomalies 501 

may also contribute to increase precipitation and air uplift over the Gulf of Guinea coast (see Fig. 11a, where 502 

KHI200/850 regression anomalies are shown in contours). Over the Sahel, the underlying dynamics of the leading 503 

precipitation pattern in Fig. 11b are harder to interpret. On the one side, a positive EM forcing tends to produce 504 

dryer conditions over this region (Losada et al. 2010; 2012). On the other side, an ENSO- forcing is known to 505 

increase air uplift and precipitation over West Africa (cf. Fig. 3). Thus, Sahelian results in Figs. 11b-c seem to be 506 

a blend of the aforementioned dynamical mechanisms. This is supported by the strong anticorrelation value (-507 

0.38; 90% confidence interval) obtained between the Atlantic (EM-WAM; Section 4.4) and Pacific (ENSO-508 

WAM; Section 4.2) expansion coefficients U (cf. Table S3). 509 

(B) GCMs & GCMs-RCA4 510 

The ensemble map of SST leading patterns obtained from individual GCMs is shown in Fig. 12a. The figure 511 

shows a robust SST warming constrained over the eastern equatorial Atlantic (scf 36%), and a well-defined 512 

ENSO- pattern over the Pacific. The outcome for individual models is provided in Fig. S13. In general, individual 513 

SST maps tend to agree in a warming pattern over the equatorial Atlantic together with statistically significant 514 

colder SSTs in the Pacific (5 out of 8 cases; cf. Table S3). However, the area covered by the Atlantic SST 515 

anomalies differs notably among models (compare CNRM-CM5 and EC-EARTH; Figs. S13d and S13g). Such 516 

limitations may be due to the systematic SST errors in the tropical Atlantic simulated by GCMs, which hamper 517 

the representation of EM variability (Fig. S1d; Richter and Xie 2008; Xu et al. 2014).  518 

Regarding the precipitation heterogeneous map, the model ensemble reveals a quite robust pattern of increased 519 

precipitation over the eastern equatorial Atlantic, and decreased over Gambia/Senegal (Fig. 12c). Specifically, 520 

this precipitation pattern is exceptionally similar to the one obtained in Fig. 1b (GCM ensemble bias). This is not 521 

accidental, as in both figures a SST warming in the equatorial Atlantic (caused by either natural variability or 522 

GCM errors) leads to a southward shift of the ITCZ. The precipitation pattern in Fig. 12c is also highly consistent 523 



19 
 

with Fig. 12e, where DIV200/850, MF850 and SHUM850 ensemble anomalies are regressed on expansion 524 

coefficient U. Over the Gulf of Guinea (Fig. 12e), the combined effect of EM+ (via enhanced moisture supply 525 

and air uplift) and ENSO- (enhanced air uplift) is consistent with positive precipitation anomalies (Fig. 12c). 526 

These anomalies are, however, shifted to the south in GCMs compared to observations due GCM SST biases. 527 

Over the Sahel, ensemble GCM precipitation is clearly reduced compared to observations (cf. Figs. 12c and 11b). 528 

One possibility to this behavior is that the influence of ENSO- is not strong enough over the Sahel to counteract 529 

the negative precipitation anomalies associated with a positive EM (Losada et al. 2010). This hypothesis is in line 530 

with our results in Section 4.2, where the ENSO influence on Sahelian precipitation is shown to be weaker in 531 

GCMs than in observations. To complement GCM ensembles, individual regression maps of precipitation and 532 

additional fields (KHI200/850, DIV200/850, MF850 etc.) are provided in Fig. S13. Overall, results are diverse 533 

among GCM simulations, being CNRM-CM5 a clear example of unrealistic modeling (Fig. S13e-f). 534 

Subsequently, GCM-RCA4 results are provided in Fig. 12 (right column). The ensemble SST homogeneous map 535 

returns a very similar pattern to that obtained from GCMs alone (compare Figs. 12a-b). Again, concurrent EM+ 536 

and ENSO- SST anomalies are present in the individual regression maps (Fig. S14). Apart from CNRM-537 

CM5/RCA4 (Fig. S14d), the Atlantic EM SST variability is reasonably captured by the individual GCM-RCA4 538 

simulations. In Fig. 12d the ensemble of GCM-RCA4 precipitation maps is provided. The pattern looks even more 539 

dipolar than the obtained for GCMs alone, with stronger/broader negative precipitation anomalies over the Sahel 540 

(compare Figs. 12c-d). According to Figs. 12e-f, the Sahelian dryer conditions in GCMs-RCA4 are linked to 541 

increased air subsidence (compare DIV200/850 anomalies in colors). These results also agree with our findings 542 

in section 4.2, where it is shown that the ENSO influence on Sahel precipitation via air subsidence is absent in 543 

GCMs-RCA4 (cf. Fig. 4). Therefore, ensemble regression patterns in Figs. 12d,f depict a more dipolar structure, 544 

typically associated with an isolated positive EM forcing (Losada et al. 2010; 2012). For completeness, regression 545 

maps of individual GCM-RCA4 simulations are also available in Fig. S14. 546 

(C) MPI & MPI-RCMs 547 

Next, the EM-WAM teleconnection is assessed on MPI (Fig. 13 - left column) and the ensemble of MPI-RCM 548 

runs (Fig. 13 - right column). The homogeneous SST map for MPI (Fig. 13a) reveals that the leading SST mode 549 

is EM+ (scf 32%). Although EM+ is accompanied by simultaneous negative SST anomalies over the Tropical 550 

Pacific, the latter are very weak (colors) and do not appear to influence the large-circulation over West Africa 551 

(KHI200/850 - contours). As a consequence, the associated precipitation heterogeneous map is markedly dipolar, 552 

with negative rainfall anomalies over the Sahel and positive in the Gulf of Guinea (Fig. 13c). Regression anomalies 553 



20 
 

of DIV200/850, MF850 and SHUM850 on U are also consistent with a nearly isolated forcing of EM+ (Fig. 13e). 554 

First, because enhanced humidity and vertical motion of air are present over the Gulf of Guinea. Second, because 555 

stronger air subsidence and dryer conditions are situated over the Sahel. These anomalies are consistent with 556 

inhibited moisture flux from the equatorial Atlantic towards continental northern Africa (cf. arrows in Fig. 13e). 557 

Subsequently, the outcome of MPI-RCM simulations is provided in Fig. 13 (right column). As expected, the 558 

ensemble homogeneous SST map is almost identical to that obtained from MPI alone (compare Figs. 13a-b). 559 

Regarding the individual simulations, in 3 out of 4 cases the SST anomalies over the central Pacific are not 560 

statistically significant (Fig. S15). As a consequence, ensemble regression anomalies in Figs. 13d (precipitation) 561 

and 13f (DIV200/850, SHUM850) are also markedly dipolar between the Gulf of Guinea and Sahel. Nevertheless, 562 

for the MPI-RCM ensemble, the anomalies appear attenuated over the Sahel area (compare Figs. 13c,e and 13d,f). 563 

The explanation to this behavior is not trivial, especially with the methodology used in this article. Thus, it can 564 

only be cautiously conjectured. On the one hand, the MPI/CCLM4 simulation seems to capture some statistically 565 

significant ENSO- anomalies accompanying EM+ in the MCA analysis (Fig. S15a). In this simulation, both 566 

regressed precipitation (Fig. S15b) and DIV200/850 (Fig. S15c) appear enhanced over the western Sahel, an 567 

aspect not observed in the rest of MPI-RCM simulations. Therefore, CCLM4 seems able, for some unknown 568 

reason, to amplify the ENSO influence on the WAM compared with the driving GCM model (MPI; Figs. 13c,e). 569 

On the other hand, dryer conditions over the Sahel area are found in the regression map of MPI (Fig. 13e - 570 

contours) compared with the MPI-RCM ensemble (Fig. 13f). Thus, a better representation of fine-scale processes 571 

of the Sahel area (e.g., soil-atmosphere interactions, land use etc.), which may be absent in the MPI run (Steiner 572 

et al. 2009; Domínguez et al. 2010; Paeth et al. 2011), could also help to improve simulations of SLP/moisture 573 

gradients/precipitation and lead to these results. 574 

To confirm the hypothesis above, several additional sensitivity experiments should be carried out using different 575 

GCM/RCM simulations. In particular, it would be interesting to calculate the MCAs setting as predictor an 576 

isolated EM+ SST pattern. However, such analysis is far beyond the scope of this paper and might be limited by 577 

the occurrence of isolated EM+ anomalies in observations (our study period spans only 26 years). Hence, it is left 578 

out for future research. 579 

Finally, no RCM added value maps are provided for the EM-WAM teleconnection. First, because the contrasting 580 

leading SST modes obtained in observations and some model simulations are associated with different 581 

teleconnection mechanisms (Figs. 11-13). Thus, it makes no sense to compare precipitation anomalies influenced 582 



21 
 

by ENSO (e.g., observations) with those which are not (e.g., MPI; cf. Table S3). Second, because the underlying 583 

dynamics are so mixed between EM and ENSO that assessing causality is not possible at this point. 584 

 585 

5. Conclusions and discussion 586 

In this article, the ability of GCMs and RCMs to represent the West African Monsoon rainfall regime and its most 587 

prominent interannual teleconnections is analyzed. For this purpose, 8 General Circulation Model (GCM) 588 

historical simulations from CMIP5 and 11 Regional Climate Model (RCM) simulations from CORDEX-Africa 589 

are considered (Giorgi et al. 2009; Nikulin et al. 2012; Taylor et al. 2012). To account for uncertainties in RCM 590 

performance and lateral boundary conditions, different GCM/RCM experiment blocks are analyzed (cf. Table 1). 591 

First, a set of 8 GCM simulations (a.k.a., GCMs) regionalized on SMHI-RCA4 (GCMs-RCA4). Second, a 592 

simulation of MPI-ESM-LR (a.k.a. MPI) driving 4 different RCMs (CCLM4-8, REMO, RCA4 and CRCM5; 593 

a.k.a. MPI-RCMs). Observational datasets are also utilized to characterize the West African Monsoon 594 

(WAM)/Sea Surface Temperature (SST) teleconnections. The period chosen for the analysis is the peak season of 595 

the WAM, July-August-September (JAS), along 26 consecutive years (1979-2004). 596 

Firstly, seasonal biases of mean and standard deviation precipitation fields are characterized (Figs. 1-2). In 597 

accordance with previous studies, GCMs and regional simulations have trouble to simulate the northward 598 

migration of the Intertropical Convergence Zone (Nikulin et al. 2012; Dosio et al. 2015). This is caused by Atlantic 599 

equatorial SST biases present in GCMs (Fig. S1d). In addition, bias improvement in regional simulations is highly 600 

sensitive to GCM performance. For instance, MPI-RCM simulations are not able to improve precipitation biases 601 

over large areas of West Africa. This could be related to the relative good performance of MPI itself (Figs. 1f and 602 

2) and the fact that convective precipitation is parameterized both in GCMs and RCMs. The opposite is found for 603 

GCM and GCM-RCA4 simulations (Figs. 1e and 2). These results may be different in convection-permitting or 604 

cloud-resolving simulations (Randall et al. 2003; Prein et al. 2015). 605 

Secondly, the ability of GCMs and RCMs to simulate the most prominent interannual teleconnection patterns of 606 

the WAM is assessed. These are, following Rodríguez-Fonseca et al. (2011; 2015): (i) El Niño-Southern 607 

Oscillation (ENSO); (ii) the eastern Mediterranean (MED); and (iii) the Atlantic Equatorial Mode (EM). With this 608 

aim, a Maximum Covariance Analysis (MCA) is performed using as predictor JAS SST anomalies from: (i) the 609 

Equatorial Pacific [110ºE-80ºW, 20ºS-20ºN]; (ii) the Mediterranean [0º-40ºE, 30ºN-45ºN]; and (iii) the equatorial 610 

Atlantic [60ºW-20ºE, 20ºS-5ºN], respectively. West African precipitation anomalies over the area [20ºW-30ºE, 611 

0º-20ºN] are considered in all analyses as predictand. For the calculations, the SST based Statistical Seasonal 612 



22 
 

Forecast model (S4CAST v.2.0), recently developed by Suárez-Moreno and Rodríguez-Fonseca (2015), is 613 

utilized. A summary of GCM/RCM performance in representing the most prominent WAM teleconnections is 614 

provided below: 615 

1. ENSO-WAM teleconnection: In observations, during a positive ENSO event, the anomalous large-scale 616 

circulation induces an intensification of air subsidence over West Africa (Fig. 3; the opposite holds for 617 

ENSO-; Joly and Voldoire 2009). Thus, precipitation is reduced all along the corridor 0º-20ºN (Fig. 3b). 618 

The ENSO signal on air subsidence/precipitation is essentially captured by GCMs (Fig. 4 – left column). 619 

However, compared to observations, the simulated anomalies appear shifted to the south due to GCM 620 

SST biases, and the influence on Sahelian precipitation is weak. Regional simulations (GCM-RCA4 and 621 

MPI-RCMs) reveal that the ENSO impact near the equatorial Atlantic is well replicated, even improved 622 

over local areas, compared with GCMs. However, over most of West Africa [20ºW-30ºE, 5ºN-20ºN], the 623 

ENSO influence on air subsidence and precipitation is clearly reduced after regionalization (cf. Figs. 4-624 

5). Hence, RCMs do not appear to be a skillful tool to improve GCM simulations of the ENSO-WAM 625 

teleconnection (Fig. 6). 626 

2. MED-WAM teleconnection: Sahelian low-level humidity and precipitation are substantially increased 627 

when the Mediterranean waters are significantly warmer (Fontaine et al. 2010). As shown in 628 

observations, a warmer Mediterranean is associated with negative SLP anomalies over Northern Africa 629 

(Saharan Heat Low) and a strong North to South SLP gradient over the Sahel. These conditions promote 630 

moisture advection from the Mediterranean (northern branch) and the Gulf of Guinea (southern branch) 631 

towards the Sahel (Fig. 7). GCMs have trouble to simulate the strength and location of the Saharan Heat 632 

Low. As a consequence, the moisture advection branches impacting the Sahel in observations (Fig. 7c) 633 

are misrepresented in GCMs (Figs. 8e and 9e). Particularly, the moisture advection branch from the Gulf 634 

of Guinea is the worst simulated. After regionalization, the pressure gradient between the Gulf of Guinea 635 

and Sahara is better characterized. As a result, the simulation of southwesterly humidity inflow and 636 

precipitation over the Sahel is improved (Figs. 8f and 9f). In this case, RCMs appear to be a skillful tool 637 

to improve GCM simulations of the MED-WAM teleconnection (Fig. 10). 638 

3. EM-WAM teleconnection: An isolated EM+ SST pattern is known to produce a dipole of precipitation 639 

anomalies over West Africa, with positive values in the Gulf of Guinea and negative over the Sahel 640 

(Losada et al. 2010). However, due to the period chosen in this study (1979-2004), simultaneous EM+ 641 

and ENSO- SST anomalies arise as the leading homogeneous MCA mode from observations (Rodríguez-642 



23 
 

Fonseca et al. 2009). Due to the combined dynamics of both SST forcings, the heterogeneous 643 

precipitation map is no longer a precipitation dipole, but a uniform mode of positive rainfall anomalies 644 

extending from the Gulf of Guinea towards the Western Sahel (Fig. 11b; Losada et al. 2010). Very similar 645 

SST modes are obtained when MCA is performed on GCMs and GCMs-RCA4 (Figs. 12-13). However, 646 

as explained above, GCMs tend to attenuate the ENSO influence on Sahelian precipitation compared to 647 

observations, and the signal gets practically removed after regionalization. Therefore, the homogeneous 648 

rainfall pattern from observations (Fig. 11b) returns a much more marked dipolar shape as far as GCM 649 

(Fig. 12c) and GCM-RCA4 (Fig. 12d) simulations are utilized. Unfortunately, for MPI and MPI-RCMs, 650 

the leading co-variability SST modes no longer present statistically significant ENSO- anomalies 651 

accompanying EM- (Figs. 13 and S14). As a result, the leading precipitation modes resemble a dipole 652 

(Figs. 13c-d) which, by construction, is not logical to compare with observations. Due to the mixed 653 

ENSO and EM dynamics in these experiments, no robust conclusions can be determined. 654 

The novel results in this article will help to select the most appropriate RCM simulations to study WAM 655 

teleconnections. This outcome is based on the following findings: (1) It has been shown that the ENSO-WAM 656 

teleconnection is depreciated in RCMs due to their inability to propagate the ENSO impact on air subsidence 657 

along their domain. This result is in line with Boulard et al. (2013), who found over South Africa similar problems 658 

with the ENSO teleconnection and attributed these deficiencies to the lateral atmospheric forcing. In this context, 659 

the use of spectral nudging to impose the large-scale atmospheric variability within the regional domain might 660 

potentially help to improve RCM simulations of the ENSO-WAM teleconnection. Future analyses on RCM 661 

architecture, lateral nesting (e.g., Davies-type) and performance should keep the focus on this issue. Since our 662 

results also apply for the so-called “ENSO-WAM kelvin wave teleconnection mechanism”, forthcoming research 663 

on this topic might also consider additional mechanisms proposed in the literature (Joly and Voldoire 2009); (2) 664 

The influence of the Mediterranean on the WAM is much better reproduced by RCMs compared to GCMs. 665 

Particularly, RCMs improve the representation of the large-scale pressure gradient between the Gulf of Guinea 666 

and Sahara, and moisture advection over the Sahel. In this case, a great part of the Mediterranean Sea is included 667 

inside the CORDEX-Africa domain (Fig. S1a) and the SST anomalies provided by the forcing GCM are prescribed 668 

along the RCM surface. Thus, the adjacent Mediterranean SST forcing appears to notably improve RCM 669 

performance. Although for a different region, these results are also in line with Boulard et al. (2013), who found 670 

that regional SST forcing over adjacent oceans favored realistic rainfall anomalies over South Africa. Future 671 

analyses on this topic should consider the MENA-CORDEX domain, which includes northern Africa, southern 672 



24 
 

Europe and the whole Arabian Peninsula (Bucchignani et al. 2015). In this context, several studies have just 673 

pointed out the importance of a warmer Mediterranean on the recent recovery from the Sahelian drought (Evan et 674 

al. 2015; Park et al. 2016); (3) Regarding the EM-WAM teleconnection, the observed mixed EM and ENSO 675 

dynamics and the existence of lateral-atmospheric and surface-prescribed SST forcings in the RCM simulations 676 

preclude to infer any robust conclusion. The ENSO imprint on this teleconnection could potentially be removed 677 

through statistical methods. However, due to the short period considered, the event to event differences and the 678 

non-linear interactions between ENSO and other processes (e.g., volcanic eruptions, anthropogenic influence, 679 

etc.) these methods result troublesome (Brönnimann 2007). Instead, sensitivity simulations using as predictor an 680 

isolated EM mode could be carried out. Much longer observational datasets ought to be considered for this 681 

challenge.  682 

Finally, results from GCMs vs. GCMs-RCA4 (lateral boundary conditions) and MPI vs. MPI-RCMs (RCM 683 

performance) provide very similar outcomes for the ENSO-WAM and MED-WAM teleconnections. These results 684 

are consistent with the main conclusion of this study: the ability of RCMs to represent WAM teleconnections 685 

appears to be highly sensitive to the regional domain boundaries and the way the external forcings are prescribed 686 

(lateral-atmospheric vs. surface-SST). For a more comprehensive assessment on this topic, additional SST forcing 687 

patterns affecting the WAM could be considered in the future (e.g., the Indian ocean and the Tropical North 688 

Atlantic; Lu and Delworth 2005; Chung and Ramathan 2006). 689 

 690 

Acknowledgments 691 

We thank the National Centers for Environmental Prediction (NCEP)/National Center for Atmospheric Research 692 

(NCAR) and the National Aeronautics and Space Administration (NASA) for the GPCP and MERRA datasets, 693 

respectively. We also thank the European Centre for Medium-Range Weather Forecasts and the Met Office 694 

Hadley Centre for the ERA-Interim and HadISST databases. We are indebted to the Coupled Model Inter-695 

comparison Project Phase 5 (CMIP5), Coordinated Regional Climate Downscaling Experiment (CORDEX) and 696 

involved institutions for providing the GCM/RCM simulations used in this study. We also thank the Earth System 697 

Grid Federation (ESGF) for making these simulations available. This study has been supported by the European 698 

Commission’s research project PREFACE (EU/FP7 2007-2013; ref. 603521). Iñigo Gómara is also supported by 699 

the Spanish Ministry of Economy and Competitiveness (“Juan de la Cierva-Formación” contract; FJCI-2015-700 

23874). Finally, we would like to thank the two anonymous reviewers, whose pertinent comments and suggestions 701 

have contributed to improve this manuscript. 702 



25 
 

References 703 

Adeniyi MO, Dilau, KA (2016) Assessing the link between Atlantic Niño 1 and drought over West Africa using 704 

CORDEX regional climate models. Theor Appl Climatol. doi:10.1007/s00704-016-2018-0 705 

Akinsanola AA, Ogunjobi KO, Gbode IE, Ajayi VO (2015) Assessing the Capabilities of Three Regional Climate 706 

Models over CORDEX Africa in Simulating West African Summer Monsoon Precipitation. Advances in 707 

Meteorology 2015: 935431. doi:10.1155/2015/935431 708 

Biasutti M, Giannini A (2006) Robust Sahel drying in response to late 20th century forcings. Geophys Res Lett 709 

33: L11706. doi:10.1029/2006GL026067 710 

Boulard D, Pohl B, Crétat J, Vigaud N (2013) Downscaling large-scale climate variability using a regional climate 711 

model: the case of ENSO over Southern Africa. Clim Dyn 40: 1141-1168. doi:10.1007/s00382-012-1400-6 712 

Bretherton CS, Smith C, Wallace JM (1992) An intercomparison of methods for finding coupled patterns in 713 

climate data. J Clim 5: 541–560 714 

Brönnimann S (2007) Impact of El Niño–Southern Oscillation on European climate. Rev Geophys 45: RG3003. 715 

doi:10.1029/2006RG000199 716 

Bucchignani E, Mercogliano P, Rianna G, Panitz HJ (2015) Analysis of ERA - Interim driven COSMO-CLM 717 

simulations over Middle East - North Africa domain at different spatial resolutions. Int J Climatol 36: 3346-718 

3369. doi:10.1002/joc.4559 719 

Castro CL, Pielke RA, Leoncini G (2005) Dynamical downscaling: an assessment of value added using a regional 720 

climate model. J Geophys Res 110: D05,108. doi:10.1029/2004JD004721 721 

Cherry S (1997) Some comments on singular value decomposition analysis. J Clim 10:1759–1761 722 

Chung CE, Ramanathan V (2006) Weakening of North Indian SST gradients and the monsoon rainfall in India 723 

and the Sahel. J Clim 19: 2036-2045. 724 

Cook KH, Vizy EK (2006) Coupled model simulations of the West African monsoon system: Twentieth- and 725 

twenty first-century simulations. J Clim 19: 3681–3703. 726 

Cook KH (2008) Mysteries of Sahel droughts. Nat Geosci 1: 647–648. 727 

Dee DP, Uppala SM, Simmons AJ, Berrisford P, Poli P, Kobayashi S, Andrae U, Balmaseda MA, Balsamo G, 728 

Bauer P, Bechtold P, Beljaars ACM, van de Berg L, Bidlot J, Bormann N, Delsol C, Dragani R, Fuentes M, 729 

Geer AJ, Haimberger L, Healy SB, Hersbach H, Hólm EV, Isaksen L, Kållberg P, Köhler M, Matricardi M, 730 

McNally AP, Monge-Sanz BM, Morcrette JJ, Park BK, Peubey C, de Rosnay P, Tavolato C, Thépaut JN, 731 



26 
 

Vitart F (2011) The ERA-Interim reanalysis: configuration and performance of the data assimilation system. 732 

Q J R Meteorol Soc 137: 553–597. doi:10.1002/qj.828 733 

Diallo I, Sylla MB, Camara M, Gaye AT (2013) Interannual variability of rainfall over the Sahel based on multiple 734 

regional climate models simulations. Theor Appl Climatol 113: 351-362. doi:10.1007/s00704-012-0791-y 735 

Diasso U, Abiodun BJ (2015) Drought modes in West Africa and how well CORDEX RCMs simulate them. 736 

Theor Appl Climatol. DOI 10.1007/s00704-015-1705-6 737 

Diatta S, Fink AH (2014) Statistical relationship between remote climate indices and West African monsoon 738 

variability. Int J Climatol 34: 3348–3367. doi:10.1002/joc.3912 739 

Di Luca A, de Elia R, Laprise R (2013) Potential for small scale added value of RCM’s downscaled climate 740 

change signal. Clim Dyn 40: 601–618. 741 

Dilley M, Chen RS, Deichmann U, Lerner-Lam AL, Arnold M (2005) Natural disaster hotspots: a global risk 742 

analysis. International Bank for Reconstruction and Development/the World Bank and Columbia University, 743 

Washington, DC, http://www.ldeo.columbia.edu/chrr/research/hotspots 744 

Domínguez M, Gaertner M, De Rosnay P, Losada T (2010) A regional climate model simulation over West Africa: 745 

parameterization tests and analysis of land-surface fields. Clim Dyn 35: 249−265. 746 

Dosio A, Panitz HJ, Schubert-Frisius M, Lüthi D (2015) Dynamical downscaling of CMIP5 global circulation 747 

models over CORDEX-Africa with COSMO-CLM: evaluation over the present climate and analysis of the 748 

added value. Clim Dyn 44: 2637–2661. doi:10.1007/s00382-014-2262-x 749 

Endris HS, Lennard C, Hewitson B, Dosio A, Nikulin G, Panitz HJ (2016) Teleconnection responses in multi-750 

GCM driven CORDEX RCMs over Eastern Africa. Clim Dyn 46: 2821. doi:10.1007/s00382-015-2734-7 751 

Evan AT, Flamant C, Lavaysse C, Kocha C, Saci A (2015) Water vapor-forced greenhouse warming over the 752 

Sahara Desert and the recent recovery from the Sahelian drought. J Clim 28: 108-123. 10.1175/JCLI-D-14-753 

00039.1 754 

Favre A, Philippon N, Pohl B, Kalognomou EA, Lennard C, Hewitson B, Nikulin G, Dosio A, Panitz HJ, Cerezo-755 

Mota (2016) Clim Dyn 46: 1799-1818. doi:10.1007/s00382-015-2677-z 756 

Flaounas E, Bastin S, Janicot S (2011) Regional climate modelling of the 2006 West African monsoon: sensitivity 757 

to convection and planetary boundary layer parameterisation using WRF. Clim Dyn 36:1083-1105. 758 

doi:10.1007/s00382-010-0785-3 759 

Folland CK, Palmer TN, Parker DE (1986) Sahel rainfall and worldwide sea temperatures, 1901–1985. Nature 760 

320: 602–607. doi:10.1038/320602a0 761 

http://www.ldeo.columbia.edu/chrr/research/hotspots


27 
 

Fontaine B, Janicot S (1996) Sea surface temperature fields associated with West African rainfall anomaly types. 762 

J Clim 9: 2935-2940. 763 

Fontaine B, Trasaska S, Janicot S (1998) Evolution of the relationship between near global and Atlantic SST mode 764 

and the rainy season in West Africa: statistical analyses and sensitivity experiments. Clim Dyn 14: 353–368. 765 

Fontaine B., Roucou P, Sivarajan S, Gervois S, Chauvin F, Rodríguez de Fonseca B, Ruti P, Janicot S (2010) 766 

Impacts of Warm and Cold situations in the Mediterranean Basins on the West African monsoon: observed 767 

connection patterns (1979-2006) and climate simulations. Clim Dyn 35: 95-114. 768 

Fontaine B, Gaetani M, Ullmann A, Roucou P (2011) Time evolution of observed July–September sea surface 769 

temperature‐ Sahel climate teleconnection with removed quasi‐ global effect (1900–2008). J Geophys Res 770 

116: D04105. doi:10.1029/2010JD014843 771 

Gaetani M, Fontaine B, Roucou P, Baldi M (2010) Influence of the Mediterranean Sea on the West African 772 

monsoon: Intraseasonal variability in numerical simulations. J Geophys Res 115: D24115. 773 

doi:10.1029/2010JD014436. 774 

Gbobaniyi E, Sarr A, Sylla MB, Diallo I, Lennard C, Dosio A, Dhiédiou A, Kamga A, Klutse NAB, Hewitson B, 775 

Nikulin G, Lamptey B (2014) Climatology, annual cycle and interannual variability of precipitation and 776 

temperature in CORDEX simulations over West Africa. Int J Climatol 34: 2241–2257. doi:10.1002/joc.3834 777 

Giorgi F, Mearns LO (1999) Introduction to special section: Regional Climate Modeling Revisited. J Geophys 778 

Res 104: 6335–6352. doi:10.1029/98JD02072 779 

Giorgi F, Jones C, Asrar GR (2009) Addressing climate information needs at the regional level: The CORDEX 780 

framework. WMO Bull 58: 175–183. 781 

Gómara I, Rodríguez-Fonseca B, Zurita-Gotor P, Ulbrich S, Pinto JG (2016) Abrupt transitions in the NAO 782 

control of explosive North Atlantic cyclone development. Clim Dyn 47: 3091-3111. doi:10.1007/s00382-016-783 

3015-9 784 

Hoerling M, Hurrell J, Eischeid J, Phillips A (2006) Detection and attribution of twentieth-century northern and 785 

southern African Rainfall Change. J Clim 19(16): 3989–4008 786 

Hourdin F, Musat I, Guichard F, Ruti PM, Favot F, Filiberti MA, Pham M, Grandpeix JY, Polcher J, Marquet P, 787 

Boone A, Lafore JP, Redelsperger JL, Dellaquila A, Losada DT, Khadre TA, Gallee H (2010) AMMA-model 788 

intercomparison project. Bull Am Meteorol Soc 91:95–104. doi:10.1175/2009BAMS2791.1 789 

Huffman GJ, Adler RF, Bolvin DT, Gu G (2009) Improving the global precipitation record: GPCP Version 2.1. 790 

Geophys Res Lett 36: L17808. 791 



28 
 

IPCC AR5 (2014) Climate Change 2014: Synthesis Report. Contribution of Working Groups I, II and III to the 792 

Fifth Assessment Report of the Intergovernmental Panel on Climate Change [Core Writing Team, R.K. 793 

Pachauri and L.A. Meyer (eds.)]. IPCC, Geneva, Switzerland, 151 pp. 794 

Janicot, S (1992) Spatio-temporal variability of West African rainfall. Part I: Regionalization and typings. J Clim 795 

5: 489-497. 796 

Janicot S, Moron V, Fontaine B (1996) Sahel drought and ENSO dynamics. Geophys Res Lett 23(5): 515–518. 797 

doi:10.1029/96GL00246 798 

Janicot S, Harzallah A, Fontaine B, Moron V (1998) West African monsoon dynamics and eastern equatorial 799 

Atlantic and Pacific SST anomalies (1970–1988). J Clim 11: 1874–1882. doi: 10.1175/1520-0442-11.8.1874. 800 

Janicot, S, Trzaska S, Poccard I (2001) Summer Sahel-ENSO teleconnection and decadal time scale SST 801 

variations. Clim Dyn 18: 303–320. 802 

Janicot S, Caniaux G, Chauvin F, de Coëtlogon G, Fontaine B, Hall N, Kiladis G, Lafore JP, Lavaysse C, Lavender 803 

SL, Leroux S, Marteau R, Mounier F, Philippon N, Roehrig R, Sultan B, Taylor CM (2011) Intraseasonal 804 

variability of the West African monsoon. Atmos Sci Lett 12: 58–66. doi:10.1002/asl.280 805 

Joly M, Voldoire A, Douville H, Terray P, Royer JF (2007) African monsoon teleconnections with tropical SSTs: 806 

Validation and evolution in a set of IPCC4 simulations. Clim Dyn 29: 1-20. doi:10.1007/s00382-006-0215-8. 807 

Joly M, Voldoire A (2009) Influence of ENSO on the West African monsoon: Temporal aspects and atmospheric 808 

processes. J Clim 22: 3193–3210. 809 

Jones CG, Giorgi F, Asrar G (2011) The Coordinated Regional Downscaling Experiment: CORDEX; An 810 

international downscaling link to CMIP5. CLIVAR Exchanges, International CLIVAR Project Office, No. 56, 811 

Southampton, United Kingdom, 34–40. [Available online at http://www. 812 

clivar.org/sites/default/files/imported/publications/exchanges/Exchanges_56.pdf] 813 

Jung T, Ferranti L, Tompkins AM (2006) Response to the Summer of 2003 Mediterranean SST Anomalies over 814 

Europe and Africa. J Clim 19: 5439–5454. 815 

Laprise R, de Elía R, Caya D, Biner S, Lucas-Picher P, Diaconescu E, Leduc M, Alexandru A, Separovic L (2008) 816 

Challenging some tenets of regional climate modelling. Meteor Atmos Phys. Special Issue on Regional 817 

Climate Studies 20: 3–22. 818 

Laprise R, Hernández-Díaz L, Tete K, Sushama L, Šeparović L, Martynov A, Winger K, Valin M (2013) Climate 819 

projections over CORDEX Africa domain using the fifth-generation Canadian Regional Climate Model 820 

(CRCM5). Clim Dyn 41: 3219. doi:10.1007/s00382-012-1651-2 821 

http://dx.doi.org/10.1175/1520-0442-11.8.1874
http://www/


29 
 

Laurent H, D’Amato N, Lebel T (1998) How important is the contribution of the mesoscale convective complexes 822 

to the Sahelian rainfall? Physics and Chemistry of the Earth 23: 629–633. 823 

Lavaysse C, Flamant C, Janicot S (2010a) Regional scale convection patterns during strong and weak phases of 824 

the Saharan heat low. Atmos Sci Lett 11: 255–264. DOI:10.1002/asl.284. 825 

Lavaysse C, Flamant C, Janicot S, Knippertz P (2010b) Links between African easterly waves, midlatitude 826 

circulation and intraseasonal pulsations of the West African heat low. Q J R Meteorol Soc 136: 141-158. 827 

DOI:10.1002/qj.555 828 

Le Barbé L, Lebel T (1997) Rainfall climatology of the HAPEX Sahel region during the years 1950–1990. Journal 829 

of Hydrology 188-189: 43–73. 830 

Li G, Xie SP (2012) Origins of tropical-wide SST biases in CMIP multi-model ensembles. Geophys Res Lett 39: 831 

L22703. doi:10.1029/2012GL053777 832 

López-Parages J, Rodríguez-Fonseca B (2012) Multidecadal modulation of El Niño influence on the Euro-833 

Mediterranean rainfall. Geophys Res Lett 39: L02704. doi:10.1029/2011GL050049 834 

Lorenz EN (1956) Empirical orthogonal functions and statistical weather prediction. Science Report 1, Statistical 835 

Forecasting Project, Department of Meteorology, MIT (NTIS AD 110268), 49 pp. 836 

Losada T, Rodríguez-Fonseca B, Janicot S, Gervois S, Chauvin F, Ruti P (2010) A multi-model approach to the 837 

Atlantic Equatorial mode: impact on the West African monsoon. Clim Dyn 35(1): 29–43. 838 

Losada T, Rodríguez-Fonseca B, Mohino E, Bader J, Janicot S, Mechoso CR (2012) Tropical SST and Sahel 839 

rainfall: A non-stationary relationship. Geophys Res Lett 39: L12705. doi:10.1029/2012GL052423 840 

Lu J, Delworth TL (2005) Oceanic forcing of the late 20th century Sahel drought. Geophys Res Lett 32: L22706. 841 

doi:10.1029/2005GL023316. 842 

Meque A, Abiodun BJ (2015) Simulating the link between ENSO and summer drought in Southern Africa using 843 

regional climate models. Clim Dyn 44: 1881-1900. doi:10.1007/s00382-014-2143-3 844 

Mesinger F, Brill K, Chuang HY, Di Mego G, Rogers E (2002) Limited area predictability: can ‘‘upscaling’’ also 845 

take place? Tech. Rep. 32: 5.30–5.31, Research Activities in Atmospheric and Oceanic Modelling, WMO, 846 

CAS/JSC WGNE, Geneva 847 

Mohino E, Janicot S, Bader J (2011a) Sahel rainfall and decadal to multi-decadal sea surface temperature 848 

variability. AGCM intercomparison. Clim Dyn 37: 1707-1725. 849 



30 
 

Mohino E, Rodríguez-Fonseca B, Mechoso CR, Gervois S, Ruti P, Chauvin F (2011b) Impacts of the tropical 850 

Pacific/Indian Oceans on the seasonal cycle of the West African monsoon. J Clim 24: 3878–3891. doi: 851 

10.1175/2011JCLI3988.1 852 

Mohino E, Rodríguez-Fonseca B, Losada T, Gervois S, Janicot S, Bader J, Ruti P, Chauvin F (2011c) Changes in 853 

the interannual SST-forced signals on West African rainfall. AGCM intercomparison. Clim Dyn 37: 1707-854 

1725. 855 

Moron V, Ward MN (1998) ENSO teleconnections with climate variability in the European and African sectors. 856 

Weather 53: 287–295. 857 

Nikulin G, Jones C, Giorgi F, Asrar G, Büchner M, Cerezo-Mota R, Christensen OB, Déqué M, Fernández J, 858 

Hänsler A, van Meijgaard E, Samuelsson P, Sylla MB, Sushama L (2012) Precipitation climatology in an 859 

ensemble of CORDEX-Africa regional climate simulations. J Clim 25: 6057–6078. 860 

Paeth H, Hall NMJ, Gaertner MA, Alonso MD, Moumouni S, Polcher J, Ruti PM, Fink AH, Gosset M, Lebel T, 861 

Gaye AT, Rowell DP, Moufouma-Okia W, Jacob D, Rockel B, Giorgi F, Rummukainen M (2011) Progress 862 

in regional downscaling of west African precipitation. Atmos Sci Lett 12: 75–82. doi:10.1002/asl.306 863 

Palmer TN (1986) Influence of the Atlantic, Pacific and Indian Oceans on Sahel rainfall. Nature 320: 251–253. 864 

doi:10.1038/322251a0 865 

Park JY, Bader J, Matei, D (2016) Anthropogenic Mediterranean warming essential driver for present and future 866 

Sahel rainfall. Nature Climate Change 6: 941-945. doi:10.1038/nclimate3065 867 

Parry ML, Canziani OF, Palutikof JP, van der Linden PJ, Hanson CE (2007) Climate Change 2007: Impacts, 868 

Adaptation and Vulnerability. Contribution of Working Group II to the Fourth Assessment Report of the 869 

Intergovernmental Panel on Climate Change, Cambridge University Press, Cambridge, UK, 2007. 870 

Polo I, Rodríguez-Fonseca B, Losada T, García-Serrano J (2008) Tropical atlantic variability modes (1979-2002)-871 

part I: time evolving SST modes related to West African rainfall. J Clim 21: 6457-6475. 872 

Prein AF, Langhans W, Fosser G, Ferrone A, Ban N, Goergen K, Keller M, Tölle M, Gutjahr O, Feser F, Brisson 873 

E, Kollet S, Schmidli J, van Lipzig NPM, Leung R (2015) A review on regional convection-permitting climate 874 

modeling: Demonstrations, prospects, and challenges. Rev Geophys 53: 323-361. 875 

doi:10.1002/2014RG000475. 876 

Raible CC, Lehner F, Gonzalez-Rouco JF, Fernández-Donado L (2014) Changing correlation structures of the 877 

Northern Hemisphere atmospheric circulation from 1000 to 2100 AD. Climate of the Past 10: 537-550. 878 

doi:10.5194/cp-10-537-2014. 879 



31 
 

Randall DA, Khairoutdinov MF, Arakawa A, Grabowski WW (2003) Breaking the cloud parameterization 880 

deadlock. Bull Amer Meteor Soc 84: 1547-1564. 881 

Rayner NA, Parker DE, Horton EB, Folland CK, Alexander LV, Rowell DP, Kent EC, Kaplan A (2003) Global 882 

analyses of sea surface temperature, sea ice, and night marine air temperature since the late nineteenth century. 883 

J Geophys Res 108: 4407. doi:10.1029/2002JD002670 884 

Richter I, Xie SP (2008) On the origin of equatorial Atlantic biases in coupled general circulation models. Clim 885 

Dyn 31: 587–598. 886 

Richter I, Xie SP, Wittenberg AT, Masumoto Y (2012) Tropical Atlantic biases and their relation to surface wind 887 

stress and terrestrial precipitation. Clim Dyn 38: 985-1001. 888 

Rienecker MM, Suarez MJ, Gelaro R, Todling R, Bacmeister J, Liu E, Bosilovich MG, Schubert SD, Takacs L, 889 

Kim GK, Bloom S, Chen J, Collins D, Conaty A, da Silva A, Gu W, Joiner J, Koster RD, Lucchesi R, Molod 890 

A, Owens T, Pawson S, Pegion P, Redder CR, Reichle R, Robertson FR, Ruddick AG, Sienkiewicz M, 891 

Woollen J (2011) MERRA: NASA's Modern-Era Retrospective Analysis for Research and Applications. J 892 

Clim 24: 3624-3648. doi:10.1175/JCLI-D-11-00015.1. 893 

Rodríguez-Fonseca B, Polo I, García-Serrano J, Losada T, Mohino E, Mechoso CR, Kucharski F (2009) Are 894 

Atlantic Niños enhancing Pacific ENSO events in recent decades? Geophys Res Lett 36: L20705. DOI: 895 

10.1029/2009GL040048. 896 

Rodríguez-Fonseca B, Janicot S, Mohino E, Losada T, Bader J, Caminade C, Voldoire A (2011) Interannual and 897 

decadal SST- forced responses of the West African monsoon. Atmos Sci Lett 12: 67-74. 898 

Rodríguez-Fonseca B, Mohino E, Mechoso CR, Caminade C, Biasutti M, Gaetani M, García-Serrano J, Vizy EK, 899 

Cook K, Xue Y, Polo I, Losada T, Druyan L, Fontaine B, Bader J, Doblas-Reyes FJ, Goddard L, Janicot S, 900 

Arribas A, Lau W, Colman A, Vellinga M, Rowell DP, Kucharski F, Voldoire A (2015) Variability and 901 

Predictability of West African Droughts: A Review on the Role of Sea Surface Temperature Anomalies. J 902 

Clim 28: 4034-4060. DOI: http://dx.doi.org/10.1175/JCLI-D-14-00130.1 903 

Rodríguez-Fonseca B, Suárez-Moreno R, Ayarzagüena B, López-Parages J, Gómara I, Villamayor J, Mohino E, 904 

Losada T, Castaño-Tierno A (2016) A Review of ENSO Influence on the North Atlantic. A Non-Stationary 905 

Signal. Atmosphere 7: 87. doi:10.3390/atmos7070087 906 

Rowell DP, Folland CK, Maskel K, Owen JA, Ward MN (1995) Variability of the summer rainfall over tropical 907 

North Africa (1906–92): observations and modeling. Q J R Meteorol Soc 121: 669–704. 908 

doi:10.1002/qj.49712152311 909 



32 
 

Rowell DP (2001) Teleconnections between the tropical Pacific and the Sahel. Q J R Meteorol Soc 127: 1683-910 

1706. 911 

Rowell DP (2003) The impact of Mediterranean SSTs on the Sahelian rainfall season. J Clim 16: 849-862. 912 

Semazzi FHM, Mehta V, Sud YC (1988) An investigation of the relationship between sub−Saharan rainfall and 913 

global sea surface temperatures. Atmosphere-Ocean 26: 118–138. 914 

Steiner AL, Pal JS, Rauscher SA, Bell JL, Diffenbaugh NS, Boone S, Sloan LC, Giorgi F (2009) Land surface 915 

coupling in regional climate simulations of the West African monsoon. Clim Dyn 33(6): 869-892. doi: 916 

10.1007/s00382-009-0543-6 917 

Suárez-Moreno R, Rodríguez-Fonseca B (2015) S4CAST v2.0: sea surface temperature based statistical seasonal 918 

forecast model. Geosci Model Dev 8: 3639-3658. doi:10.5194/gmd-8-3639-2015 919 

Sultan B, Janicot S (2000) Abrupt shift of the ITCZ over West Africa and intra-seasonal variability. Geophys Res 920 

Lett 27: 3353–3356. doi: 10.1029/1999GL011285 921 

Sylla MB, Coppola E, Mariotti L, Giorfi F, Ruti PM, Dell’Aquila A, Bi X (2010) Multiyear simulation of the 922 

African climate using a regional climate model (RegCM3) with the high resolution ERA-interim reanalysis. 923 

Clim Dyn 35: 231–247. 924 

Taylor KE, Stouffer RJ, Meehl GA (2012) An Overview of CMIP5 and the experiment design. Bull Am Meteorol 925 

Soc 93: 485-498. doi:10.1175/BAMS-D-11-00094.1 926 

Trenberth KE (1997) The Definition of El Niño. Bull Am Meteorol Soc 78: 2771-2777. 927 

Veljovic K, Rajkovic B, Fennessy MJ, Altshuler EL, Mesinger F (2010) Regional climate modeling: should one 928 

attempt improving on the large scales? Lateral boundary condition scheme: any impact? Meteorol Z 19(3): 929 

237–246. 930 

Vizy EK, Cook KH (2001) Mechanisms by which Gulf of Guinea and eastern North Atlantic sea surface 931 

temperature anomalies can influence African rainfall. J Clim 14: 795-821. 932 

Widmann M (2005) One-dimensional CCA and SVD, and their relationship to regression maps. J Clim 18: 2785-933 

2792. 934 

Wilks DS (2006) Statistical methods in the Atmospheric Sciences, Second Edition, Elsevier Inc., ISBN 13: 978-935 

0-12-751966-1 936 

Xu Z, Chang P, Richter I, Kim W, Tang G (2014) Diagnosing southeast tropical Atlantic SST and ocean 937 

circulation biases in the CMIP5 ensemble. Clim Dyn 43: 3123-3145. doi:10.1007/s00382-014-2247-9 938 



33 
 

Yaka P, Sultan B, Broutin H, Janicot S, Philippon S, Fourquet N (2008) Relationships between climate and year-939 

to-year variability in meningitis outbreaks: a case study in Burkina Faso and Niger. Int J Health Geogr 7: 34. 940 

doi:10.1186/1476-072X-7-34 941 

Yin X, Gruber A (2010) Validation of the abrupt change in GPCP precipitation in the Congo River basin. Int J 942 

Climatol 30: 110-119. 943 

Zebiak SE (1993) Air–sea interaction in the equatorial Atlantic region. J Clim 6:1567-1586.  944 



34 
 

Figure Captions: 945 

 946 

Fig. 1: (a) Mean (shadings; mm day-1) and standard deviation (contours; mm day-1) of GPCP JAS seasonal 947 

precipitation (1979-2004). Rectangles denote West-Africa North (WA-N), West-Africa South (WA-S) and the 948 

whole West Africa (WA) regions. (b) Ensemble biases of GCMs vs. GPCP (1979-2004). Differences in 949 

mean/standard deviation are given in colors/contours (mm day-1). +/x symbols denote areas where all models 950 

depict the same sign in mean/standard deviation biases. (c) Same as (b) but for the ensemble of GCM-RCA4 951 

simulations. (d) Same as (b) but for the ensemble of MPI-RCM simulations. (e) Ensemble added value in 952 

mean/standard deviation of GCM-RCA4 simulations in colors/contours (mm2 day-2). 75% of members giving the 953 

same added value sign in mean/standard deviation are marked with +/x. (f) Same as (e) but for MPI-RCMs. 954 

 955 

Fig. 2: (a) Biases in mean precipitation values averaged over West Africa – North [10ºW-10ºE, 9ºN-15ºN] with 956 

respect to observations (GPCP – horizontal offset; mm day-1). Blue bars correspond to GCM biases. Yellow bars 957 

correspond to GCM-RCA4 biases. Red bars correspond to MPI-RCM biases. The horizontal red line provides the 958 

climatological value from MERRA. (b) Same as (a) but for standard deviation biases over West-Africa – North 959 

(mm day-1). (c)-(d) Same as (a)-(b) but for the West Africa – South region [10ºW-10ºE, 5ºN-9ºN]. (e)-(f) Same as 960 

(a)-(b) but for the whole West Africa region [20ºW-30ºE, 0ºN-20ºN]. 961 

 962 

Fig. 3: (a) JAS SST anomalies (HadISST) regressed on expansion coefficient U in colors (homogeneous map: K 963 

std-1). 90% confidence interval in stippling (Monte Carlo test: 1000 random iterations). Differences in velocity 964 

potential anomalies at 200 hPa with respect to those at 850 hPa (KHI200/850; ERA-Interim) regressed on U in 965 

contours (every 0.8 106 m2 s-1). (b) JAS precipitation anomalies (GPCP) regressed on U in colors (heterogeneous 966 

map; mm day-1 std-1). 90% confidence interval in stippling. (c) Differences in wind divergence anomalies at 200 967 

hPa with respect to those at 850 hPa (DIV200/850; ERA-Interim) regressed on U in colors (10-6 s-1 std-1). 90% 968 

confidence interval in stippling. Differences in velocity potential anomalies at 200 hPa with respect to those at 969 

850 hPa (KHI200/850; every 0.4 106 m2 s-1) regressed on U in contours. Expansion coefficient U is obtained from 970 

a Maximum Covariance Analysis. Squared covariance fraction provided in %. Predictor: SST [110ºE-80ºW, 20ºS-971 

20ºN; dashed rectangle in (a)]. Predictand: PCP [20ºW-30ºE, 0º-20ºN; dashed rectangle in (b)]. Period: 1979-972 

2004. 973 

 974 



35 
 

Fig. 4: (a)-(c)-(e) Same as Fig. 3 but for the ensemble mean of GCMs (stippling - 7/8 models with same sign on 975 

regression in SST, rainfall and DIV200/850). (b)-(d)-(f) Same as (a)-(c)-(e) but for the ensemble mean of GCMs-976 

RCA4. 977 

 978 

Fig. 5: (a)-(c)-(e) Same as Fig. 3 but for the MPI simulation. (b)-(d)-(f) Same as (a)-(c)-(e) but for the ensemble 979 

mean of MPI-RCMs. 4/4 models with same sign on regression: shadings in (b), stippling in (d)-(f). 980 

 981 

Fig 6: (a) Ensemble added value (AV) map (colors, in mm2 day-2 std-2) of the ENSO-WAM teleconnection for 982 

precipitation, calculated as the average of AV values from each individual model. Positive values: areas where 983 

GCMs-RCA4 improve GCM simulations of the ENSO-WAM teleconnection (considering observations as basis). 984 

Red/Blue contours: number of simulations of each experiment with positive AVs over each grid point. (b) Same 985 

as (a) but for MPI-RCM simulations. 986 

 987 

Fig 7: Same as Fig. 3 but using as predictor Mediterranean SST anomalies [0º-40ºE, 30ºN-45ºN]. No KHI200/850 988 

regression anomalies are plotted. Regression maps on (c) are based on anomalous specific humidity at 850 hPa 989 

(SHUM850 - colors, in Kg Kg-1 std-1; 90% confidence level in green circles), sea level pressure (SLP - contours, 990 

hPa std-1) and moisture flux at 850 hPa (MF850 - arrows, Kg Kg-1 m s-1 std-1). 991 

 992 

Fig. 8: (a)-(c)-(e) Same as Fig. 7 but for the ensemble mean of GCMs (stippling/circles - 7/8 models with same 993 

sign on regression in SST, precipitation and SHUM850). (b)-(d)-(f) Same as (a)-(c)-(e) but for the ensemble mean 994 

of GCMs-RCA4. 995 

 996 

Fig. 9: (a)-(c)-(e) Same as Fig.7 but for the MPI simulation. (b)-(d)-(f) Same as (a)-(c)-(e) but for the ensemble 997 

mean of MPI-RCMs. 4/4 models with same sign on regression: shadings in (b), stippling/circles in (d)-(f) 998 

 999 

Fig. 10: Same as Fig. 6 but for the Mediterranean-WAM teleconnection.  1000 

 1001 

Fig 11: Same as Fig. 3 but using as predictor Atlantic SST anomalies [60ºW-20ºE, 20ºS-5ºN]. Regression maps 1002 

on (c) are based on anomalous specific humidity at 850 hPa (SHUM850 - contours, in Kg Kg-1 std-1), moisture 1003 



36 
 

flux at 850 hPa (MF850 - vectors, Kg Kg-1 m s-1 std-1) and wind divergence difference between 200/850 hPa 1004 

(DIV200/850 - colors, 10-6 s-1 std-1; 90% confidence interval in white circles). 1005 

 1006 

Fig. 12: (a)-(c)-(e) Same as Fig. 11 but for the ensemble mean of GCMs (stippling/circles - 7/8 models with same 1007 

sign on regression in SST, precipitation and DIV200/850). (b)-(d)-(f) Same as (a)-(c)-(e) but for the ensemble 1008 

mean of GCMs-RCA4. 1009 

 1010 

Fig. 13: (a)-(c)-(e) Same as Fig. 11 but for the MPI simulation. (b)-(d)-(f) Same as (a)-(c)-(e) but for the ensemble 1011 

mean of MPI-RCMs. 4/4 models with same sign on regression: shadings in (b), stippling/circles in (d)-(f). 1012 

 1013 



(b)         Ensemble Bias: GCMs

SEASONAL PRECIPITATION BIASES

(a)       Mean Precipitation: GPCP

(d)      Ensemble Bias: MPI-RCMs(c)    Ensemble bias: GCMs-RCA4

(f)    Ens. Added Value: MPI-RCMs(e)  Ens. Added Value: GCMs-RCA4

Figure1



(a)                              Mean

AVERAGE BIAS OVER WEST AFRICA - NORTH

(b)                       Standard Deviation

AVERAGE BIAS OVER WEST AFRICA - SOUTH

(c)                              Mean (d)                     Standard Deviation

(e)                              Mean (f)                      Standard Deviation

AVERAGE BIAS OVER ALL WEST AFRICA

Figure2



MCA LEADING MODE: OBSERVATIONS

(a)                        Predictor: HadISST (41%)

(b)             Predictand: GPCP (c)    U Regression Maps (ERA-Int.)

Figure3



(d) Predictand: Ens. GCMs-RCA4 PCP

MCA LEADING MODES: GCMs & GCMs-RCA4

(c)     Predictand: Ens. GCMs PCP

(f) Ens. GCMs-RCA4: U Regression Maps(e)  Ens. GCMs: U Regression Maps

(b)   Predictor: Ens. GCMs-RCA4 SST (33%)(a)           Predictor: Ens. GCMs SST (32%)

Figure4



(d)   Predictand: Ens. MPI-RCMs PCP

MCA LEADING MODES: MPI & MPI-RCMs

(c)           Predictand: MPI PCP

(f) Ens. MPI-RCMs: U Regression Maps(e)        MPI: U Regression Maps

(b)      Predictor: Ens. MPI-RCMs SST (31%)(a)               Predictor: MPI SST (33%)

Figure5



(b)                     MPI-RCMs

ENSO-WAM TELECONNECTION: ENSEMBLE ADDED VALUES

(a)                  GCMs-RCA4

Figure6



MCA LEADING MODE: OBSERVATIONS

(a)                        Predictor: HadISST (40%)

(b)             Predictand: GPCP (c)    U Regression Maps (ERA-Int.)

Figure7



(d) Predictand: Ens. GCMs-RCA4 PCP

MCA LEADING MODES: GCMs & GCMs-RCA4

(c)     Predictand: Ens. GCMs PCP

(f) Ens. GCMs-RCA4: U Regression Maps(e)  Ens. GCMs: U Regression Maps

(b)   Predictor: Ens. GCMs-RCA4 SST (33%)(a)           Predictor: Ens. GCMs SST (32%)

Figure8



(d)   Predictand: Ens. MPI-RCMs PCP

MCA LEADING MODES: MPI & MPI-RCMs

(c)           Predictand: MPI PCP

(f) Ens. MPI-RCMs: U Regression Maps(e)        MPI: U Regression Maps

(b)      Predictor: Ens. MPI-RCMs SST (26%)(a)               Predictor: MPI SST (29%)

Figure9



(b)                     MPI-RCMs

MED-WAM TELECONNECTION: ENSEMBLE ADDED VALUES

(a)                  GCMs-RCA4

Figure10



MCA LEADING MODE: OBSERVATIONS

(a)                        Predictor: HadISST (41%)

(b)             Predictand: GPCP (c)    U Regression Maps (ERA-Int.)

Figure11



(d) Predictand: Ens. GCMs-RCA4 PCP

MCA LEADING MODES: GCMs & GCMs-RCA4

(c)     Predictand: Ens. GCMs PCP

(f) Ens. GCMs-RCA4: U Regression Maps(e)  Ens. GCMs: U Regression Maps

(b)   Predictor: Ens. GCMs-RCA4 SST (34%)(a)           Predictor: Ens. GCMs SST (36%)

Figure12



(d)   Predictand: Ens. MPI-RCMs PCP

MCA LEADING MODES: MPI & MPI-RCMs

(c)           Predictand: MPI PCP

(f) Ens. MPI-RCMs: U Regression Maps(e)        MPI: U Regression Maps

(b)      Predictor: Ens. MPI-RCMs SST (28%)(a)               Predictor: MPI SST (32%)

Figure13



Tables: 

 

Table 1: CMIP5/CORDEX-Africa historical simulations matrix. 

 

Historical CMIP5 GCMs 

CORDEX-Africa RCMs 

Matrix 
C

a
n

E
S

M
2

 

C
C

M
A

-C
a

n
a
d

a
 

C
N

R
M

-C
M

5
 

C
N

R
M

-F
ra

n
c
e 

E
C

-E
A

R
T

H
 (

r1
2

) 

E
C

M
W

F
-E

u
ro

p
ea

n
 

 
G

F
D

L
-E

S
M

2
M

 

N
O

A
A

 G
F

D
L

-U
S

A
 

H
a

d
G

E
M

2
-E

S
 

M
et

 O
ff

ic
e
-U

K
 

 
M

IR
O

C
5

 

J
A

M
S

T
E

C
-J

a
p
a

n
 

M
P

I-
E

S
M

-L
R

 

M
P

I-
G

e
rm

a
n

y 

N
o

rE
S

M
1

-M
 

N
C

C
-N

o
rw

a
y 

CLMcom-CCLM4-8 

CCLM community-International 

        

CSC-REMO 

MPI-Germany 

        

SMHI-RCA4 

SMHI-Sweden 

        

UQAM-CRCM5 

UQAM-Canada 

        

 

CORDEX-Africa RCMs (Giorgi et al. 2009) and CMIP5 GCMs (Taylor et al. 2012) used. Shaded cells indicate 

selected GCM-RCM combinations. All data were retrieved from the Earth System Grid Federation (ESGF) portals, 

regridded to the CanESM2 horizontal resolution (coarser one; 2.8º x 2.8º) and selected for the period July-August-

September 1979-2004. The Institute ID of each model is provided in bold italics. 

 

Manuscript Tables



1 
 

 

ELECTRONIC SUPPLEMENTARY MATERIAL FOR:  

 

‘Impact of dynamical regionalization on precipitation biases and 

teleconnections over West Africa’ 

Iñigo Gómara1,2,3*, Elsa Mohino1, Teresa Losada1, Marta Domínguez 1,2, Roberto Suárez-Moreno1,2 and Belén Rodríguez-Fonseca 1,2 

 

1 Dpto. Geofísica y Meteorología, Universidad Complutense de Madrid, Madrid, Spain 

 

2 Instituto de Geociencias (IGEO), UCM, CSIC, Madrid, Spain 

 

3 CEIGRAM, Universidad Politécnica de Madrid, Madrid, Spain 

 

 

 

 

 

 

 

 

Submitted to 

Climate Dynamics 

11 April 2017 

Revised on 16 August 2017 

  

                                                           
*Correspondence to: Iñigo Gómara, Dpto. Geofísica y Meteorología, Universidad Complutense de Madrid, Facultad de CC. Físicas, 

Ciudad Universitaria s/n. 28040 Madrid, Spain. E-mail: i.gomara@ucm.es 



2 
 

CORDEX DOMAIN, PRECIPITATION OBSERVATIONS and GCM SST biases 
 

(a)                          CORDEX Africa domain  

 

(b)  MERRA - mean precipitation 

 
(c)   Difference: MERRA - GPCP 

 
(d)                                                                       GCM SST biases 

 
Fig. S1: (a) CORDEX Africa simulation domain [24.64ºW-60.28ºE, 45.76ºS-42.24ºN] and 1-degree resolution elevation (in m). 

Source: NCAR Terrain Base (TBASE) dataset. (b) Mean (shadings; mm day-1) and standard deviation (contours; mm day-1) of 

MERRA JAS seasonal precipitation (1979-2004). Rectangles denote West-Africa North (WA-N), West-Africa South (WA-S) and the 

whole West Africa (WA) regions. (c) Difference between MERRA and GPCP databases in mean (shadings) and standard deviation 

(contours) for JAS 1979-2004. (d) Ensemble SST bias of GCMs used in this article compared with HadISST (JAS 1979-2004; shadings 

in K). Areas where SST biases are of the same sign in all GCM simulations (8/8) in contours. 

 

 

  



3 
 

 
Model Biases: General Circulation Models (GCMs) 

(a) CanESM2 

 

(b) CNRM-CM5 

 

(c) EC-EARTH (r12) 

 

(d) GFDL-ESM2M 

 
(e) HadGEM2-ES 

 

(f) MIROC5 

 

(g) MPI-ESM-LR 

 

(h) NorESM1-M 

 
 

Model Biases: GCMs-RCA4 
(i) CanESM2/RCA4 

 

(j) CNRM-CM5/RCA4 

 

(k) EC-EARTH/RCA4 

 

(l) GFDL-ESM2M/RCA4 

 
(m) HadGEM2-ES/RCA4 

 

(n) MIROC5/RCA4 

 

(o) MPI-ESM-LR/RCA4 

 

(p) NorESM1-M/RCA4 

 
 

Model Biases: MPI-RCMs 
(q) MPI/CLMcom-CCLM4-8 

 

(r) MPI/CSC-REMO 

 

(s) MPI/SMHI-RCA4 

 

(t) MPI/UQAM-CRCM5 

 
Fig. S2: (a)-(h) Biases of GCM historical simulations (8 models in total, cf. Table 1) with respect to the GPCP database (period 1979-

2004). Differences in mean values are given in colors (units - mm day-1). Differences in standard deviation are provided in contours 

(units - mm day-1). Rectangles denote West-Africa North (WA-N), West-Africa South (WA-S) and the whole West Africa (WA) 

regions, respectively. (i)-(p) Same as (a)-(h) but for GCM-RCA4 simulations. (q)-(t) Same as (a)-(h) but for MPI-RCM simulations. 



4 
 

 

Added Value maps: GCMs-RCA4 
(a) CanESM2/RCA4 

 

(b) CNRM-CM5/RCA4 

 

(c) EC-EARTH/RCA4 

 

(d) GFDL-ESM2M/RCA4 

 
(e) HadGEM2-ES/RCA4 

 

(f) MIROC5/RCA4 

 

(g) MPI-ESM-LR/RCA4 

 

(h) NorESM1-M/RCA4 

 
 

Added Value maps: MPI-RCMs 
(i) MPI/CLMcom-CCLM4-8 

 

(j) MPI/CSC-REMO 

 

(k) MPI/SMHI-RCA4 

 

(l) MPI/UQAM-CRCM5 

 
Fig. S3: (a)-(h) Added value maps for the individual model members (RCA4 vs. GCMs) in representing JAS mean precipitation (colors, 

in mm2 day-2) and standard deviation (contours, in mm2 day-2) (1979-2004). Positive RCM added values in mean are highlighted with 

stippling. (i)-(l) Same as (a)-(h) but for MPI-RCM simulations. Rectangles denote West-Africa North (WA-N), West-Africa South 

(WA-S) and the whole West Africa (WA) regions, respectively. 

  



5 
 

SST Modes 
Eq. Pacific Mediterranean Eq. Atlantic 

(a) EOF1 - Exp. Var. 58% 

 

(b) EOF1 - Exp. Var. 52% 

 

(c) EOF1 - Exp. Var. 64% 

 
Precipitation Modes over West Africa 

(d) EOF1 - Exp. Var. 30% 

 

(e) EOF2 - Exp. Var. 25% 

 

(f) EOF3 - Exp. Var. 8% 

 
Fig. S4: (a)-(c) Leading Empirical Orthogonal Functions (EOFs) of JAS SST anomalies (linear trend removed; in K std-1) over the 

Equatorial Pacific [110ºE-80ºW, 20ºS-20ºN], Mediterranean [0º-40ºE, 30ºN-45ºN] and Equatorial Atlantic [60ºW-20ºE, 20ºS-5ºN], 

respectively. (d)-(f) Three leading EOFs of JAS precipitation anomalies (linear trend removed; in mm day-1 std-1) over West Africa 

[20ºW-30ºE, 0º-20ºN]. Explained variance provided in % in all figure panels. Period: 1979-2004. Databases: HadISST/GPCP. 

 

  



6 
 

MCA LEADING MODES: Individual GCM simulations 
 

PREDICTOR: JAS SST 
 [110E-80W, 20S-20N] 

PREDICTAND: JAS PCP  
[20W-30E, 0-20N] 

U Regression Maps 

(a) CanESM2 - 30% 
 

 

(b) CanESM2 

 

(c) CanESM2 

 
(d) CNRM-CM5 - 25% 
 

 

(e) CNRM-CM5 

 

(f) CNRM-CM5 

 
(g) EC-EARTH (r12) - 26% 
 

 

(h) EC-EARTH (r12) 

 

(i) EC-EARTH (r12) 

 
(j) GFDL-ESM2M - 31% 
 

 

(k) GFDL-ESM2M 

 

(l) GFDL-ESM2M 

 
Continued… 
 
 
 
 

 
 
 
 
 

 
 
 
 
 



7 
 

(m) HadGEM2-ES - 28% 
 

 

(n) HadGEM2-ES 

 

(o) HadGEM2-ES 

 
(p) MIROC5 - 45% 
 

 

(q) MIROC5 

 

(r) MIROC5 

 
(s) MPI-ESM-LR - 33% 
 

 

(t) MPI-ESM-LR 

 

(u) MPI-ESM-LR 

 
(v) NorESM1-M - 36% 
 

 

(w) NorESM1-M 

 

(x) NorESM1-M 

 
Fig. S5: Left column: JAS SST anomalies (from individual GCM simulations) regressed on expansion coefficient U in colors 

(homogeneous map: K std-1). 90% confidence interval in stippling (Monte Carlo test: 1000 random iterations). Difference in velocity 

potential anomaly at 200 hPa with respect to 850 hPa (KHI200/850; GCMs) regressed on U in contours (every 0.8 106 m2 s-1). Central 

column: JAS precipitation anomalies (from individual GCMs) regressed on U in colors (heterogeneous map; mm day-1 std-1). 90% 

confidence interval in stippling. Right column: Difference in wind divergence anomaly (DIV200/850; from individual GCMs) regressed 

on U in colors (10-6 s-1 std-1). 90% confidence interval in stippling. Difference in velocity potential anomaly regressed on U in contours 

(same as in left column but every 0.4 106 m2 s-1). Expansion coefficient U is obtained from a Maximum Covariance Analysis. Squared 

covariance fraction provided in %. Predictor: SST [110ºE-80ºW, 20ºS-20ºN; dashed rectangle in left column]. Predictand: PCP [20ºW-

30ºE, 0º-20ºN; dashed rectangle in central column]. Period: 1979-2004. 



8 
 

MCA LEADING MODES: Individual GCM-RCA4 simulations 
 

PREDICTOR: JAS SST 
 [110E-80W, 20S-20N] 

PREDICTAND: JAS PCP  
[20W-30E, 0-20N] 

U Regression Maps 

(a) CanESM2/RCA4 - 35% 
 

 

(b) CanESM2/RCA4 

 

(c) CanESM2/RCA4 

 
(d) CNRM-CM5/RCA4 - 21% 
 

 

(e) CNRM-CM5/RCA4 

 

(f) CNRM-CM5/RCA4 

 
(g) EC-EARTH (r12)/RCA4 - 24% 
 

 

(h) EC-EARTH (r12)/RCA4 

 

(i) EC-EARTH (r12)/RCA4 

 
(j) GFDL-ESM2M/RCA4 - 42% 
 

 

(k) GFDL-ESM2M/RCA4 

 

(l) GFDL-ESM2M/RCA4 

 
Continued… 
 
 
 
 

 
 
 
 
 

 
 
 
 
 



9 
 

 
(m) HadGEM2-ES/RCA4 - 26% 
 

 

 
(n) HadGEM2-ES/RCA4 

 

 
(o) HadGEM2-ES/RCA4 

 
(p) MIROC5/RCA4 - 54% 
 

 

(q) MIROC5/RCA4 

 

(r) MIROC5/RCA4 

 
(s) MPI-ESM-LR/RCA4 - 33% 
 

 

(t) MPI-ESM-LR/RCA4 

 

(u) MPI-ESM-LR/RCA4 

 
(v) NorESM1-M/RCA4 - 33% 
 

 

(w) NorESM1-M/RCA4 

 

(x) NorESM1-M/RCA4 

 
Fig. S6: Same as Fig. S5 but for GCM-RCA4 individual simulations. 

 

  



10 
 

MCA LEADING MODES: Individual MPI-RCM simulations 
 

PREDICTOR: JAS SST 
 [110E-80W, 20S-20N] 

PREDICTAND: JAS PCP  
[20W-30E, 0-20N] 

U Regression Maps 

(a) MPI-ESM-LR/CLMcom-CCLM4-8 - 34% 
 

 

(b) MPI/CCLM4 

 

(c) MPI/CCLM4 

 
(d) MPI-ESM-LR/CSC-REMO - 30% 
 

 

(e) MPI/REMO 

 

(f) MPI/REMO 

 
(g) MPI-ESM-LR/SMHI-RCA4 - 33% 
 

 

(h) MPI/RCA4 

 

(i) MPI/RCA4 

 
(j) MPI-ESM-LR/UQAM-CRCM5 - 28% 
 

 

(k) MPI/CRCM5 

 

(l) MPI/CRCM5 

 
Fig. S7: Same as Fig. S5 but for MPI-RCM individual simulations. 

 

 



11 
 

ADDED VALUE GCMs-RCA4 - PACIFIC 
(a) CanESM2/RCA4 

 

(b) CNRM-CM5/RCA4 

 

(c) EC-EARTH (r12)/RCA4 

 

(d) GFDL-ESM2M/RCA4 

 
(e) HadGEM2-ES/RCA4 

 

(f) MIROC5/RCA4 

 

(g) MPI-ESM-LR/RCA4 

 

(h) NorESM1-M/RCA4 

 
 

ADDED VALUE MPI-RCMs - PACIFIC 
(i) MPI/CCLM4 

 

(j) MPI/CSC-REMO 

 

(k) MPI/RCA4 

 

(l) MPI/CRCM5 

 
Fig. S8: (a)-(h) Added value maps (colors, in mm2 day-2 std-2) for the individual model members (RCA4 vs. GCMs) in representing the 

ENSO-WAM teleconnection (JAS 1979-2004). Positive values are highlighted with stippling. (i)-(l) Same as (a)-(h) but for MPI-RCMs 

vs. MPI-ESM-LR. 

 

 

  



12 
 

MCA LEADING MODES: Individual GCM simulations 
 

PREDICTOR: JAS SST 
 [0W-40E, 30-45N] 

PREDICTAND: JAS PCP  
[20W-30E, 0-20N] 

U Regression Maps 

(a) CanESM2 - 29% 
 

 

(b) CanESM2 

 

(c) CanESM2 

 
(d) CNRM-CM5 - 26% 
 

 

(e) CNRM-CM5 

 

(f) CNRM-CM5 

 
(g) EC-EARTH (r12) - 30% 
 

 

(h) EC-EARTH (r12) 

 

(i) EC-EARTH (r12) 

 
(j) GFDL-ESM2M - 31% 
 

 

(k) GFDL-ESM2M 

 

(l) GFDL-ESM2M 

 
Continued… 
 
 
 

 
 
 
 

 
 
 
 



13 
 

 
 
(m) HadGEM2-ES - 24% 
 

 

 
 
(n) HadGEM2-ES 

 

 
 
(o) HadGEM2-ES 

 
(p) MIROC5 - 35% 
 

 

(q) MIROC5 

 

(r) MIROC5 

 
(s) MPI-ESM-LR - 29% 
 

 

(t) MPI-ESM-LR 

 

(u) MPI-ESM-LR 

 
(v) NorESM1-M - 49% 
 

 

(w) NorESM1-M 

 

(x) NorESM1-M 

 
Fig. S9: Left/central columns: Same as Fig. S5 but using as predictor JAS SST anomalies from the Mediterranean [0º-40ºE, 30º-45ºN]. 

No regression anomalies of KHI200/850 are provided. Right column: Anomalous specific humidity at 850 hPa (SHUM850 - colors, in 

Kg Kg-1 std-1), sea level pressure (SLP - contours, hPa std-1) and moisture flux at 850 hPa (MF850 - arrows, Kg Kg-1 m s-1 std-1) regressed 

on expansion coefficient U. 90% confidence interval in green circles (only for specific humidity anomalies).  



14 
 

MCA LEADING MODES: Individual GCM-RCA4 simulations 
 

PREDICTOR: JAS SST 
 [0-40E, 30-45N] 

PREDICTAND: JAS PCP  
[20W-30E, 0-20N] 

U Regression Maps 

(a) CanESM2/RCA4 - 27% 
 

 

(b) CanESM2/RCA4 

 

(c) CanESM2/RCA4 

 
(d) CNRM-CM5/RCA4 - 42% 
 

 

(e) CNRM-CM5/RCA4 

 

(f) CNRM-CM5/RCA4 

 
(g) EC-EARTH (r12)/RCA4 - 22% 
 

 

(h) EC-EARTH (r12)/RCA4 

 

(i) EC-EARTH (r12)/RCA4 

 
(j) GFDL-ESM2M/RCA4 - 29% 
 

 

(k) GFDL-ESM2M/RCA4 

 

(l) GFDL-ESM2M/RCA4 

 
Continued… 
 
 
 

 
 
 
 

 
 
 
 



15 
 

 
 
(m) HadGEM2-ES/RCA4 - 29% 
 

 

 
 
(n) HadGEM2-ES/RCA4 

 

 
 
(o) HadGEM2-ES/RCA4 

 
(p) MIROC5/RCA4 - 37% 
 

 

(q) MIROC5/RCA4 

 

(r) MIROC5/RCA4 

 
(s) MPI-ESM-LR/RCA4 - 30% 
 

 

(t) MPI-ESM-LR/RCA4 

 

(u) MPI-ESM-LR/RCA4 

 
(v) NorESM1-M/RCA4 - 46% 
 

 

(w) NorESM1-M/RCA4 

 

(x) NorESM1-M/RCA4 

 
Fig. S10: Same as Fig. S9 but for GCM-RCA4 individual simulations. 

  



16 
 

MCA LEADING MODES: Individual MPI-RCM simulations 
 

PREDICTOR: JAS SST 
 [0-40E, 30-45N] 

PREDICTAND: JAS PCP  
[20W-30E, 0-20N] 

U Regression Maps 

(a) MPI-ESM-LR/CLMcom-CCLM4-8 - 26% 
 

 

(b) MPI/CCLM4 

 

(c) MPI/CCLM4 

 
(d) MPI-ESM-LR/CSC-REMO - 24% 
 

 

(e) MPI/REMO 

 

(f) MPI/REMO 

 
(g) MPI-ESM-LR/SMHI-RCA4 - 30% 
 

 

(h) MPI/RCA4 

 

(i) MPI/RCA4 

 
(j) MPI-ESM-LR/UQAM-CRCM5 - 26% 
 

 

(k) MPI/CRCM5 

 

(l) MPI/CRCM5 

 
Fig. S11: Same as Fig. S9 but for MPI-RCM individual simulations. 

  



17 
 

ADDED VALUE GCMs-RCA4 - MEDITERRANEAN 
(a) CanESM2/RCA4 

 

(b) CNRM-CM5/RCA4 

 

(c) EC-EARTH (r12)/RCA4 

 

(d) GFDL-ESM2M/RCA4 

 

(e) HadGEM2-ES/RCA4 

 

(f) MIROC5/RCA4 

 

(g) MPI-ESM-LR/RCA4 

 

(h) NorESM1-M/RCA4 

 
 

ADDED VALUE MPI-RCMs - MEDITERRANEAN 
(i) MPI/CCLM4 

 

(j) MPI/CSC-REMO 

 

(k) MPI/RCA4 

 

(l) MPI/CRCM5 

 
Fig. S12: (a)-(h) Added value maps (colors, in mm2 day-2 std-2) for the individual model members (RCA4 vs. GCMs) in representing 

the MED-WAM teleconnection (JAS 1979-2004). Positive values are highlighted with stippling. (i)-(l) Same as (a)-(h) but for MPI-

RCMs vs. MPI-ESM-LR. 

 

  



18 
 

MCA LEADING MODES: Individual GCM simulations 
 

PREDICTOR: JAS SST 
 [60W-20E, 20S-5N] 

PREDICTAND: JAS PCP  
[20W-30E, 0-20N] 

U Regression Maps 

(a) CanESM2 - 36% 
 

 

(b) CanESM2 

 

(c) CanESM2 

 
(d) CNRM-CM5 - 21% 
 

 

(e) CNRM-CM5 

 

(f) CNRM-CM5 

 
(g) EC-EARTH (r12) - 28% 
 

 

(h) EC-EARTH (r12) 

 

(i) EC-EARTH (r12) 

 
(j) GFDL-ESM2M - 54% 
 

 

(k) GFDL-ESM2M 

 

(l) GFDL-ESM2M 

 
Continued… 
 
 

 
 
 

 
 
 



19 
 

 
 
 
(m) HadGEM2-ES - 40% 
 

 

 
 
 
(n) HadGEM2-ES 

 

 
 
 
(o) HadGEM2-ES 

 
(p) MIROC5 - 45% 
 

 

(q) MIROC5 

 

(r) MIROC5 

 
(s) MPI-ESM-LR - 32% 
 

 

(t) MPI-ESM-LR 

 

(u) MPI-ESM-LR 

 
(v) NorESM1-M - 32% 
 

 

(w) NorESM1-M 

 

(x) NorESM1-M 

 
Fig. S13: Left/central columns: Same as Fig. S5 but using as predictor JAS SST anomalies from the equatorial Atlantic [60ºW-20ºE, 

20ºS-5ºN]. Right column: Anomalous specific humidity at 850 hPa (SHUM850 - contours, in Kg Kg-1 std-1), moisture flux at 850 hPa 

(MF850 - arrows, in Kg Kg-1 m s-1 std-1) and difference in wind divergence at 200 hPa with respect to 850 hPa (DIV200/850 – colors, 

in 10-6 s-1 std-1) regressed on expansion coefficient U. 90% confidence interval in white circles for wind divergence anomalies. 



20 
 

MCA LEADING MODES: Individual GCM-RCA4 simulations 
 

PREDICTOR: JAS SST 
 [60W-20E, 20S-5N] 

PREDICTAND: JAS PCP  
[20W-30E, 0-20N] 

U Regression Maps 

(a) CanESM2/RCA4 - 40% 
 

 

(b) CanESM2/RCA4 

 

(c) CanESM2/RCA4 

 
(d) CNRM-CM5/RCA4 - 24% 
 

 

(e) CNRM-CM5/RCA4 

 

(f) CNRM-CM5/RCA4 

 
(g) EC-EARTH (r12)/RCA4 - 30% 
 

 

(h) EC-EARTH (r12)/RCA4 

 

(i) EC-EARTH (r12)/RCA4 

 
(j) GFDL-ESM2M/RCA4 - 54% 
 

 

(k) GFDL-ESM2M/RCA4 

 

(l) GFDL-ESM2M/RCA4 

 
Continued… 
 
 
 

 
 
 
 

 
 
 
 



21 
 

 
 
(m) HadGEM2-ES/RCA4 - 22% 
 

 

 
 
(n) HadGEM2-ES/RCA4 

 

 
 
(o) HadGEM2-ES/RCA4 

 
(p) MIROC5/RCA4 - 45% 
 

 

(q) MIROC5/RCA4 

 

(r) MIROC5/RCA4 

 
(s) MPI-ESM-LR/RCA4 - 28% 
 

 

(t) MPI-ESM-LR/RCA4 

 

(u) MPI-ESM-LR/RCA4 

 
(v) NorESM1-M/RCA4 - 26% 
 

 

(w) NorESM1-M/RCA4 

 

(x) NorESM1-M/RCA4 

 
Fig. S14: Same as Fig. S13 but for GCM-RCA4 individual simulations. 

 

 



22 
 

MCA LEADING MODES: Individual MPI-RCM simulations 
 

PREDICTOR: JAS SST 
 [60W-20E, 20S-5N] 

PREDICTAND: JAS PCP  
[20W-30E, 0-20N] 

U Regression Maps 

(a) MPI-ESM-LR/CLMcom-CCLM4-8 - 28% 
 

 

(b) MPI/CCLM4 

 

(c) MPI/CCLM4 

 
(d) MPI-ESM-LR/CSC-REMO - 26% 
 

 

(e) MPI/REMO 

 

(f) MPI/REMO 

 
(g) MPI-ESM-LR/SMHI-RCA4 - 28% 
 

 

(h) MPI/RCA4 

 

(i) MPI/RCA4 

 
(j) MPI-ESM-LR/UQAM-CRCM5 - 28% 
 

 

(k) MPI/CRCM5 

 

(l) MPI/CRCM5 

 
Fig. S15: Same as Fig. S13 but for MPI-RCM individual simulations. 

  



23 
 

 

Supplementary Tables: 

 

Table S1: Correlation values between Expansion Coefficients U and the leading Pacific, Mediterranean and Atlantic SST modes 

Correlation Equatorial Pacific Mediterranean Equatorial Atlantic 

U vs. PC1SST 0.99 0.91 0.99 

 

Correlation values between Pacific (ENSO-WAM), Mediterranean (MED-WAM) and Atlantic (EM-WAM) expansion coefficients U 

and the leading Principal Components (PC1s) of SST anomalies over the same regions (cf. EOFs from Figs. S4a-c) in observations. 

Boldface indicates 90% confidence interval (t test). 

 

Table S2: Correlation values between Expansion Coefficients V and the leading precipitation modes over West Africa 

Correlation VPAC VMED VATL 

PC1PCP-WA 0.91 -0.76 -0.81 

PC2PCP-WA 0.27 -0.51 0.50 

PC3PCP-WA 0.04 -0.03 -0.08 

 

Correlation values between Pacific (ENSO-WAM), Mediterranean (MED-WAM) and Atlantic (EM-WAM) expansion coefficients V 

and the three leading Principal Components (PCs) of JAS precipitation anomalies over West Africa (cf. EOFs from Figs. S4d-f) in 

observations. Boldface indicates 90% confidence interval (t test). 

 

Table S3: Correlation values of Pacific (ENSO) and Atlantic (EM) Expansion Coefficients U 

Correlation Observations CanESM2 CNRM EC-E GFDL HadGEM2 MIROC5 MPI NorESM1 

UPAC vs. UATL -0.38 -0.34 -0.10 -0.45 -0.50 -0.37 -0.61 -0.27 -0.44 

 Ensemble GCMs: -0.38 

 

Correlation values between Pacific (ENSO-WAM) and Atlantic (EM-WAM) expansion coefficients U in observations and GCMs. 

Boldface indicates 90% confidence interval (t test).