Generated using version 3.1.2 of the official AMS LATEX template

The present and future of the West African monsoon: a1

process-oriented assessment of CMIP5 simulations along the2

AMMA transect.3

Romain Roehrig ∗ Dominique Bouniol, Francoise Guichard

CNRM-GAME, Toulouse, France

4

Frédéric Hourdin

LMD, IPSL, Paris, France

5

Jean-Luc Redelsperger

LPO, Brest, France

6

∗Corresponding author address: Romain Roehrig, Météo-France, CNRM-GAME/GMGEC/EAC, 42 av-

enue G. Coriolis, 31057 Toulouse Cedex, France.

E-mail: romain.roehrig@meteo.fr

1

ABSTRACT7

The present comprehensive assessment of the West African monsoon in the models of the8

fifth phase of the Coupled Model Intercomparison Project (CMIP5) indicates little evolution9

since CMIP3 in terms of both biases in present-day climate and climate projections.10

The outlook for Sahel precipitation in twenty-first-century coupled simulations remains11

uncertain, as models still disagree on the sign of the trends. This contrasts with the relatively12

robust spring and summer warming of the Sahel, larger by 10 to 50% compared to the global13

warming.14

CMIP5 coupled models underestimate the monsoon decadal variability, although simula-15

tions with imposed sea surface temperatures (SSTs) succeed in capturing the recent partial16

recovery of monsoon rainfall. Coupled models still suffer of major SST biases in the equato-17

rial Atlantic, inducing a systematic southward shift of the monsoon. Because of these strong18

biases, further evaluation of the African monsoon is performed in SST-imposed CMIP5 sim-19

ulations along the AMMA meridional transect, across a range of timescales ranging from20

seasonal cycle, intraseasonal fluctuations and diurnal cycle.21

A specific emphasis is made on the comprehensive set of observational data now available22

to evaluate in depth the monsoon representation across those scales and on the promising23

usefulness of high-frequency outputs provided by some CMIP5 models at selected sites along24

the AMMA transect. Most of CMIP5 models capture many features of the African mon-25

soon with varying degrees of accuracy. In particular, the top-of-atmosphere and surface26

energy balances, in relation with the cloud cover, and the intermittence and diurnal cycle of27

precipitation, demand further work from modeling centers for achieving a reasonable realism.28

1

1. Introduction29

During the second half of the twentieth century, Africa witnessed one of the largest30

interdecadal climate signal of the recent observational records. The severe drying of the31

Sahel, which culminated in the devastating drought of 1984, plagued the region from the32

70’s to the 80’s (e.g., Nicholson 1980; Nicholson et al. 2000; Held et al. 2005). In the recent33

decade, the Sahel transitioned to a period with somewhat more abundant rainfall, suggesting34

a possible shift to a more favorable climate regime over the coming decades (Paeth and35

Hense 2004). However, at the same time, global mean temperature is increasing in response36

to increasing atmospheric greenhouse gases (GHGs), so that the evolution of Sahel rainfall37

at a range of a few decades to the end of the twenty-first century becomes urgently needed38

for developing adaptation strategies.39

Such climate projections, as well as our physical understanding of the Sahel rainfall40

variability, rely for a large part on general circulation models, characterized by a wide variety41

of complexity, going from atmosphere-only models to the more recent fully coupled Earth42

system models (ESMs). Important disagreement among those models remain with regards to43

the regional response to global warming. The models of the third phase of the Coupled Model44

Intercomparison Project (CMIP3 – Meehl et al. 2007) used for the Fourth Assessment Report45

(AR4) of the Intergovernmental Panel on Climate Change (IPCC) even disagree on the sign46

of future rainfall anomalies over the Sahel (e.g., Biasutti and Giannini 2006; Cook and Vizy47

2006; Lau et al. 2006). This disagreement remains even among models that reasonably48

simulate the twentieth-century West African climate (Cook and Vizy 2006).49

In fact, a large number of the previous generation of climate models failed in capturing50

major features of the West African climatology and variability, which damps our confidence in51

the climate projection they are able to produce. One of the reason may be linked to the high52

spatial and temporal heterogeneities of rainfall distribution across West Africa. In particular,53

in the Sahel, which lies at the northernmost extent of the West African monsoon (WAM),54

between 10◦N and 20◦N, precipitation is highly sensitive to the monsoon position. There,55

2

rainfall over the Sahel occurs mostly during the June-September period, and is supplied by56

mesoscale convective systems, often organized within synoptic disturbances such as African57

easterly waves (e.g., Kiladis et al. 2006).58

Several studies emphasized the difficulties of current coupled or atmospheric models to59

handle correctly the main WAM characteristics. Cook and Vizy (2006) showed that one60

third of the CMIP3 models do not simulate a WAM system, i.e. they do not capture prop-61

erly the summer northward migration of the InterTropical Convergence Zone (ITCZ) over62

the continent. Atmospheric regional and global models, forced by observed Sea Surface63

Temperatures (SSTs), analyzed within the framework of the AMMA-MIP1, WAMME2 and64

CORDEX3-Africa projects are generally more skillful, even though large rainfall biases re-65

main (Hourdin et al. 2010; Xue et al. 2010; Boone et al. 2010). The meridional circulation66

of the monsoon displays large biaises in coupled and atmospheric models (Cook and Vizy67

2006; Hourdin et al. 2010; Xue et al. 2010).68

In the framework of the fifth phase of CMIP (CMIP5), a new ensemble of state-of-the-69

art climate models becomes available (Taylor et al. 2012), raising several questions. Do70

they agree more on Sahel rainfall projections? Do they capture the partial rainfall recovery71

observed over the last decades? How well are they able to reproduce the main features of the72

WAM? In the following, the CMIP5 ensemble is used to partly address these questions, assess73

the results of the modeling community efforts and emphasize the challenges that remain to74

it for simulating the WAM. Our analysis indicates that over West Africa, CMIP5 models75

have not reached yet a degree of maturity which makes it possible to rely directly on them76

to anticipate climate changes and their impacts, especially with regards to rainfall.77

The present study is also motivated by the recent progresses done in the observation and78

understanding of the WAM, thanks to the first phase of the AMMA program (Redelsperger79

1The African Monsoon Multidisciplinary Analyses (AMMA) Model Intercomparison Project (Hourdin

et al. 2010).2West African Monsoon Modeling and Evaluation (Xue et al. 2010).3the Coordinated Regional climate Downscaling Experiment(Jones et al. 2011; Nikulin et al. 2012)

3

et al. 2006). The observational strategy of AMMA (Lebel et al. 2010) aims at documenting80

a meridional transect extended from the Gulf of Guinean to the Sahara desert, along the81

Greenwich meridian. Three preexisting surface-observing super-sites disposed along this82

transect were reinforced: the Upper Ouémé Valley, the Niamey site and Gourma site. This83

transect was used within the AMMA-MIP framework to evaluate imposed-SST regional and84

global climate models (Hourdin et al. 2010).85

As part of the Cloud Feedback Model Intercomparison Project (CFMIP) component of86

CMIP5, participating centers also provided output at very high frequency (when possible87

at the model time step) on a series of 119 grid points around the world, in order to better88

understand the climate model behaviors and their dependence on model formulation (Bony89

et al. 2011). Among these sites, eleven were defined in coordination with the AMMA com-90

munity along the AMMA transect. three of them correspond more specifically to the AMMA91

super-sites mentioned above. In addition, the availability of new space borne measurements92

from active sensors as part of the A-train opens the path for the establishment of global93

climatologies of the three-dimensional distribution of clouds (e.g., Bouniol et al. 2012). The94

availability of all those new datasets, new outputs at sselected sites from CMIP5, as well95

as the better understanding of some key processes at work in the WAM system provides a96

unique opportunity to evaluate more in depth the WAM representation by climate models. In97

the present work, we seek to capitalize on this AMMA legacy, and provide a process-oriented98

analysis of CMIP5 simulations.99

The paper is organized as it follows: section 2 describes the various dataset used for the100

CMIP5 model evaluation. In section 3, the long-term variability of the WAM is then assessed101

from CMIP3 to CMIP5 models. Section 4 proposes an evaluation of the representation of102

its mean state and seasonal evolution in both coupled and SST-forced simulations. Section103

5 addresses a more physical evaluation of the monsoon processes, with an emphasis on the104

intraseasonal and diurnal scales of the water and energy cycles. Finally, conclusions are105

given in section 6.106

4

2. Datasets107

a. Climate models from CMIP3 and CMIP5108

In the present work, we have considered a wide range of outputs from climate models109

which participated to CMIP3 and CMIP5. Climate change scenarios of CMIP3 (SRES4 A2)110

and of CMIP5 (RCP54.5 and RCP8.5), in comparison with historical simulations (20C3M for111

CMIP3, Historical for CMIP5) are used to assess the West African monsoon response to an112

increase of the CO2 atmospheric concentration. SST-imposed or AMIP6-type simulations are113

used to further analyze the representation of the WAM in the state-of-the-art models of the114

CMIP5 archive. Pre-industrial control runs (PiControl) with constant forcing are used for115

some CMIP5 models, to infer the decadal and interannual variability of Sahel precipitation.116

A full description of the CMIP3 framework and a comprehensive assessment of the mod-117

els can be found in Meehl et al. (2007). Integrations of 18 CMIP3 models are used here.118

They were made available to the community by the Program for Climate Model Diag-119

nosis and Intercomparison (PCMDI) through their website (www-pcmdi.llnl.gov/ipcc/120

model_documentation/ipcc_model_documentation.php), where a detailed description of121

the models can be found.122

The simulations performed as part of CMIP5 and used in the present study are listed in123

Table 1. They were made available on the Earth System Grid (ESG, http://cmip-pcmdi.124

llnl.gov/cmip5/index.html) data archive. The different types of integrations of the125

CMIP5 framework are described in Taylor et al. (2012). As we provide hereafter a more126

detailed evaluation of the CMIP5 AMIP simulations, grid information are given on the at-127

mospheric component of the models which provided this experiment in Table 2.128

4Special Report on Emissions Scenarios5Representative Concentration Pathway6Atmospheric Model Intercomparison Project

5

b. Observational datasets129

The present paper assesses several timescales in CMIP5 models, from the multidecadal130

variability to the diurnal cycle. Therefore a wide number of datasets for each targeted131

variable is used to account for these scales and for the uncertainty of observations. Their132

main characteristics are described in Table 3.133

3. Long-term variability of the WAM from CMIP3 to134

CMIP5135

a. Climate projection of temperature and precipitation over the Sahel136

The CMIP3 exercise confirmed the high probability of a significant global warming at137

global scale, as a consequence of the anthropic greenhouse gases emissions (e.g., Meehl and138

Coauthors 2007). Some robust features in the spatial distribution of the associated cli-139

mate changes were also highlighted: a stronger warming over continents than over oceans,140

a stronger warming in high latitudes than in the tropics and, in general, a tendency to141

reinforcement of rainfall contrasts in the tropics, the ITCZ becoming more rainy and the142

sub-tropical subsiding anticyclonic belts drier (e.g., Chou et al. 2009). However, a major143

issue was raised as models did not achieve any consensus for regional rainfall changes, in144

particular over the Sahel (Cook 2008).145

In CMIP5 scenarios, the dispersion remains large, with still no consensus on the sign of146

rainfall evolution, as illustrated in Fig. 1.a. As in CMIP3, Sahel precipitation changes spread147

equally around zero, with most of the simulations showing moderate changes of ±20%, to148

be compared with the 40% decrease observed between the 50-60s and 70-80s, and the +20%149

rainfall recovery in recent years.150

In CMIP3, two models predicted a decrease in precipitation greater than 40%. In con-151

trast, in the RCP8.5 CMIP5 scenario, three models simulate an increase of JAS monsoon152

6

rainfall amount by 70% to 130%. Those three models also predict a relatively weak warming153

over the Sahel. In the RCP8.5 scenario, their JAS values of (∆T2m, ∆Pr/Pr) over the154

domain [13W-10E,10N-18N] are (0.4 K, 70%) for FGOALS-g2, (2.0 K, 129%) of MIROC-155

ESM-CHEM and (2.1 K, 96%) for MIROC-ESM, while the values of all other models range156

in (4.5±1.5 K, 0±20%).157

Both effects are likely related. Reinforced rainfall should moderate the temperature in158

summer, as larger water availability increase the surface latent heat flux. Figure 1.b and c159

confirm such an impact of precipitation on surface temperature. In the dry March-April-May160

season, the Sahel warming reflects mostly an amplification of the global warming response,161

ranging between 10 and 50%, with an average value of 30%. In contrast, the JAS Sahel162

warming has much more spread than the global warming, emphasizing the role of the rainfall163

response. Three models thus predict a JAS warming much smaller than the global value.164

They all show a significant increase in Sahel rainfall.165

b. Decadal and interannual rainfall variability over the Sahel166

Sahel rainfall exhibits a large variability at decadal and interannual timescales. For167

addressing these scales, the time series of the Sahel precipitation P was decomposed into168

a decadal component P 9 and an interannual fluctuation δP , such as P = P9

+ δP . P 9 is169

defined as the 9-year running mean of the raw series. Fig. 2.a illustrates the observed raw170

and filtered time series of the CRU and CMAP datasets. The Sahelian drought is clearly171

identified after 1973, with a tendency to recovery in the recent years.172

The skill of CMIP5 models to reproduce this recent recovery is first addressed in AMIP173

simulations through the computation of mean difference between the 199-2004 and 1983–1988174

periods of the filtered rainfall P 9 (Fig. 2.a). The relative change between those two periods175

reaches 13% in the CMAP observations. AMIP simulations exhibit a large dispersion. On176

average, models capture the tendency to rainfall recovery, except four of them, which have a177

slight negative tendency. This might be partly due to internal variability as illustrated when178

7

an ensemble of AMIP simulations are performed with the IPSL-CM5A-LR model (Fig. 2.a).179

Each of the five members of the ensemble, which only differ by the specification of the initial180

atmospheric state, predicts a recovery, but with an amplitude varying from +5% to +17%.181

The overall good skill of the AMIP simulations is probably related to the monsoon re-182

sponse to the change of SSTs, consistently with the success of several atmospheric models183

forced by SSTs to reproduce the main outlines of the twentieth century Sahel rainfall (e.g.,184

Tippett and Giannini 2006; Hoerling et al. 2006).185

The standard deviation of P 9 in the pre-industrial control and historical experiments186

can be used to assess the skill of the coupled atmosphere-ocean model to reproduce the187

observed decadal variability (Fig. 2.b). In CRU observations, the standard deviation over the188

twentieth century reaches 10%. Most of the models strongly underestimate this amplitude,189

by a factor two on average, in both types of experiments, with the notable exceptions of190

IPSL-CM5B-LR, which overestimates the amplitude of the decadal variability, and of BCC-191

CSM1-1, which is close to the observed amplitude. It is also remarkable that the amplitude192

of decadal variability is consistent for each model across the two experiments, suggesting193

that decadal fluctuations in historical runs are not forced by green-house gases, aerosols or194

land-use (for models including land-use changes).195

Interannual variability of CMIP5 models is investigated based on the normalized standard196

deviation of the interannual fluctuations δP (Fig. 2.c). The observed value of 10-12% is197

consistent in CMAP and CRU observations. In the historical and pre-industrial simulations,198

all models lie between 8% and 16%, except GISS-ER-2 (7%), which underestimates, and199

BCC-CSM1-1 (20%) and IPSL-CM5B-LR (26%), which overestimate interannual variability.200

The skill of these last two models to capture a large amplitude of rainfall variations at decadal201

scales, compatible with observations, may just be a consequence of excessive year-to-year202

variability.203

8

4. The representation of the West African monsoon mean204

state from CMIP3 to CMIP5205

a. Precipitation bias in historical simulations and its relationship with surface air tempera-206

ture207

The sensitivity of the Sahelian rainfall to SSTs has important consequences on the skill208

of climate models to simulate properly the present-day mean state of the monsoon system.209

All coupled models suffer from significant and robust SST biases with respect to their AMIP210

version. Most of them systematically display strong warm biases (of several K) over upwelling211

regions, on the eastern side of tropical oceanic basins, especially in the south Atlantic (Fig.212

3). The only exception is CSIRO-Mk3-6-0, which has a global cold bias over the ocean,213

particularly strong in the south Atlantic upwelling region.214

The CMIP5 ensemble mean warm bias (ENSEMBLE in Fig. 3) peaks at more than +3215

K, contrasting with a cold bias of about 1 K in the North Atlantic. This systematic bias216

structure is also remarkably similar to the CMIP3 ensemble mean (CMIP3 in Fig. 3).217

This warm bias in the equatorial Atlantic has been shown to be partly responsible for the218

systematic southward shift of the ITCZ in coupled models (Richter and Xie 2008), associated219

with a strong reinforcement of rainfall over the Guinean coast and often a reduction over220

the Sahel, as illustrated in Fig. 3. Consistently, CSIRO-Mk3-6-0 shows an opposite coupling221

with the ocean, with less rainfall over the Guinean coast.222

The latitudinal position of the ITCZ over West Africa is quite robustly correlated with223

the intensity of the north-south temperature gradient, partly driven by SSTs in the equatorial224

Atlantic in historical CMIP3 and CMIP5 simulations (Fig. 4). AMIP simulations also follow225

within this relationship, with a smaller spread in the ITCZ position, so that an important226

role of temperature over the Sahara can be inferred. Temperature over the Sahara are further227

evaluated in the following.228

To summarize, coupled models of CMIP5, similarly to those of CMIP3, exhibit large229

9

biases in the mean position of the west African monsoon, which likely originates from the230

warm SST bias in the equatorial Atlantic. This first-order, robust and quasi-systematic,231

bias prevents any further insight into the representation of key features and processes of the232

monsoon in coupled simulations. Therefore, we focus on AMIP simulations in the following,233

which display a weaker dispersion in the ITCZ summer position over West Africa (Fig. 4).234

b. The WAM mean state in AMIP simulations235

1) Precipitation236

Figure 5 shows JAS precipitation averaged from 1979 to 2008 between 10◦W and 10◦E for237

each model and each observational dataset introduced in section 2. Even though these latter238

datasets do not cover exactly the same period, they have very similar behavior along this239

transect, the sensitivity to the exact chosen period being much smaller than the typical model240

biases (not shown). Overall, they capture the large-scale idea of a precipitation maximum241

over the continent near 10-11◦N. A few models (FGOALS-s2, HadGEM2-A, inmcm4 and242

the two IPSL-CM5 models) locate their ITCZ a bit too much to the south, near 7-8◦N. In243

contrast, few models succeed in reproducing the maximum amount of precipitation along244

the transect (8 mm day−1). Four overestimate this maximum from 1.5 to 4 mm day−1,245

GFDL-HIRAM-C180, GFDL-HIRAM-C360, CSIRO-Mk3-6-0 and MIROC5. Most of the246

others underestimate it from 1 to 5 mm day−1, so that only three models are in qualitative247

agreement with the observations: CNRM-CM5, CanAM4 and IPSL-CM5A-LR. About one248

half of the models simulate far too little rainfall over the Sahel, i.e. north of 12◦N.249

The very high-resolution runs of GFDL-HIRAM (see Table 2) capture well the monsoon250

latitudinal structure, but exhibit similar skills to other models in reproducing the amplitude251

of precipitation. Besides, little sensitivity to the passage from a 0.5◦ to a 0.25◦ resolution252

is noticed. Enhanced vertical resolution from MPI-ESM-LR (47) to MPI-ESM-MR (95)253

results in a very similar ITCZ pattern. The modification of the physical packages from IPSL-254

10

CM5A-LR to IPSL-CM5B-LR indicates a clear dependence on the model physic formulation,255

especially north of 10◦N, where rainfall is decreased by almost a factor two in IPSL-CM5B-256

LR.257

2) Temperature at 2 m and the Saharan Heat Low258

Part of the spread described in the previous section can be partly related to the meridional259

large-scale temperature gradient as already discussed in section 4.a (Fig. 4). In AMIP260

simulations, this gradient is driven at first order by the temperature in the Saharan heat low261

region, which was shown in recent studies to be a key feature of the west African monsoon262

at the seasonal (Lavaysse et al. 2009) and intraseasonal (Chauvin et al. 2010) timescales.263

During the summer, a heat low establishes a low pressure system over the Sahara desert,264

which acts to reinforce the moist monsoon flow over the Sahel. The Saharan heat low is265

also key in the maintenance of the African easterly jet in the mid troposphere (Thorncroft266

and Blackburn 1999). The strong temperature gradient associated with it is a source of267

baroclinic energy for African easterly waves, which have a relevant impact on rainfall at268

various timescales (e.g. Fink and Reiner 2003; Kiladis et al. 2006; Thorncroft and Rowell269

1998). In particular, Ruti and Dell’Aquila (2010) showed that models characterized by low270

meridional temperature gradient appears to have not enough energy to feed these synoptic271

disturbances, but that a strong gradient is not a necessary condition for arising waves.272

CMIP5 models exhibit a large spread in near-surface temperature over the Sahara, which273

reaches almost 7 K near 25◦N (Fig. 6). In fact, this spread already starts from 10◦N, over the274

Sahel, and extends up to the northern coast of Africa at 35◦N. Unfortunately the dispersion275

is as large as in observational datasets and reanalyses, which first reflects the sparse coverage276

of in-situ observations over the Sahara, and second precludes at this stage any detailed model277

evaluation there. A brief analysis of the surface energy budget over the Sahara in section278

5.b will provide further insight into the origin of this spread within the CMIP5 ensemble.279

Over the Sahel, observations and reanalyses are in better agreement. There, the spread280

11

among models is weaker, although about one third of CMIP5 models are too cold by 2-281

3 K (CCSM4, CNRM-CM5, GFDL-HIRAM-C180 and -C360, NorESM1-M), and only one282

slightly too warm by 1-2 K (MRI-CGCM3).283

3) Clouds and their radiative effect284

Bouniol et al. (2012) analyzed the cloud cover mean properties over the Sahel using the285

AMF data of Niamey. They clearly identified four cloud categories: cloud associated with286

convective systems, and low-, mid- and high-level clouds, in agreement with the earlier work287

of Slingo (1980). Using CloudSat and CALIPSO, they also documented their seasonal cycle288

at the regional scale: northward migration of deep convective clouds associated with the289

ITCZ, low-level shallow clouds over the Sahel mainly during summer and ubiquitous pres-290

ence of mid-level clouds and cirrus clouds. In addition, a 2-km-deep layer of stratocumulus291

is observed over the Gulf of Guinea. Figure 7 (first panel) presents the JAS climatologi-292

cal latitude/altitude cross-section of the mean cloud fraction, built with the four years of293

CloudSat-CALIPSO data. Note that the precipitating water phase was discarded in the294

observations.295

All the models capture to some extent the observed cloud structure (Fig. 7). The296

maximum in cloud fraction related to the deep convective systems is collocated with the297

mean ITCZ position (Fig. 6), although some models do not reproduce the observed vertical298

extent of cloud fraction. Most models include in their cloud fraction only non-precipitating299

condensed water, whereas in the observational dataset, the computed cloud fraction also300

accounts for precipitating particles, especially above the freezing level. Even dense aggregates301

found in convective anvils need about 50 minutes to fall down from the 8-km altitude to the302

freezing level at a 1 m s−1 fall speed (Bouniol et al. 2010). The apparent underestimation of303

cloud fraction in the mid-troposphere, in ACCESS1-3 or IPSL-CM5A-LR, may thus originate304

from the non-consideration of precipitating ice as cloud, whose radiative impact might be305

important (Waliser et al. 2011).306

12

The high amount of mid-level clouds between 15◦N and 30◦N is one the specificities of the307

region. None of the models manages to reproduce the observed amount, even if some of them308

(CanAM4, IPSL-CM5B-LR, MIROC5) partly capture their occurrence. The stratocumulus309

over the Gulf of Guinea are also challenging most of the models, as they are often not deep310

enough when they occur. CNRM-CM5 and CanAM4 completely miss this cloud type.311

The proper representation of these different cloud types is important for the regional312

energy budget and the associated cloud feedbacks. Figure 8 shows the cloud radiative effect313

(CRE) at the top of the atmosphere. The longwave CRE is strongly shaped by the convective314

cloud cover amount and vertical structure, and latitudinal shift of its maximum is clearly315

explained by the spread of the ITCZ JAS location (Fig. 5).316

The shortwave CRE displays two minimum associated with the stratocumulus clouds317

over the ocean and the ITCZ over the continent. The spread across the simulations in the318

shortwave CRE is larger. Most of them overestimate it over the ocean, except CNRM-CM5319

which underestimates it due to a lack of stratocumulus clouds there (Fig. 7). Further320

north, within the ITCZ, the two IPSL models and HadGEM2-A underestimate the CRE.321

IPSL-CM5A-LR shows little response to the cloud cover increase with latitude.322

Even though shortwave and longwave CRE partly compensate each other, they have dis-323

tinct latitudinal structure. The longwave CRE maximum is shifted 5◦-northward compared324

to the shortwave CRE minimum. As a result, the observed net CRE is negative south of325

the ITCZ, as the shortwave component is dominating, and turns to slightly positive north of326

the ITCZ, where the longwave component dominates. Over most of the Sahel and Sahara,327

mid-level and convective clouds have a positive net CRE. It provides there more favorable328

conditions for the development of convection than further south during the monsoon (Chou329

and Neelin 2003). Only one half of the models capture the Sahel band of positive net CRE,330

but no model reproduces accurately its meridional structure, with the right balance between331

the shortwave and longwave components. North of the ITCZ, the CRE is generally too332

small, and to the south the spread is very large.333

13

c. The seasonal cycle of the WAM334

At the seasonal timescale, the West African monsoon is characterized by a northward335

migration of the ITCZ with an abrupt climatological northward shift on 24 June on average336

(Sultan and Janicot 2000, 2003), culminating in August, and by a smoother southward337

withdrawal of the rainfall band in September and October. The monsoon onset time is338

rather consistent in the two observational datasets shown in Fig. 9.r and s. It is well339

marked by a transition between a maximum of precipitation along the Guinean coast in340

May-June and a second one centered near 12◦N in August. In-between, a minimum of341

rainfall occurs over the region as the ITCZ moves northward. Presumably because the342

monsoon is primarily forced by the annual excursion of the sun, most of models capture343

the ITCZ summer migration, however with varying degrees of accuracy. Three models344

do not reproduce the spring precipitation maximum near the Guinea coast (FGOALS-g2,345

inmcm4 and IPSL-CM5A-LR), while six of them overestimate it (CNRM-CM5, CSIRO-346

Mk3-6-0, GFDL-HIRAM-C180 and C360, MPI-ESM-LR and MR). The monsoon is almost347

inexistent in BCC-CSM1-1, very weak in FGOALS-s2 and rather weak in HadGEM2-A,348

IPSL-CM5B-LR, MPI-ESM-LR and MR and MRI-CGCM3, consistently with Fig. 5. When349

simulated, the onset occurs at approximately the correct time of the year, as in CNRM-350

CM5, CSIRO-Mk3-6-0, the two versions of GFDL-HIRAM, the two versions of MPI-ESM,351

and NorESM1-M. For three models (CNRM-CM5, CSIRO-Mk3-6-0 and FGOALS-g2), the352

1-mm isoyet reaches latitudes further north than 20◦N, which is observed neither in TRMM353

nor in GPCP. On the opposite, three models keep some significant rain over the Gulf of354

Guinea during the summer (inmcm4, IPSLCM5A-LR and IPSL-CM5B-LR).355

The annual cycle of temperature in the Sahel displays a distinctive sub-tropical structure356

with two annual maxima, which are not simply related to the two annual maxima of solar357

insolation at the top of the atmosphere (e.g., Slingo et al. 2009; Guichard et al. 2009). The358

absolute maximum occurs in spring, prior to the monsoon rainfall onset, when the soil is still359

very dry, and typically at the same time as the establishment of a humid low-level monsoon360

14

flow in the Sahel. The summer rainfall leads to enhanced surface evapotranspiration (Timouk361

et al. 2009) and to an overall cooling at the surface which is the most pronounced in August.362

A second, weaker, maximum of temperature is then observed in autumn, at the retreat of363

the monsoon flow, when the insolation is still high. The temperature then rapidly drops364

to its winter absolute minimum. The annual cycle depicted by the CRU, ERA-Interim365

and MERRA are very consistent over this region. It is noticeable that the NCEP-CFSR366

reanalysis is systematically colder by 2 K all year long except for the monsoon season. This367

perhaps surprising result is nevertheless fully consistent with recent comparisons performed368

over land (Wang et al. 2011; Bao and Zhang 2012).369

Very large spread is found among climate models. The bi-modal structure is qualitatively370

captured by most of them, but the time at which the temperature reaches its maxima varies371

by up to three months in the simulations. In some models, the too quick onset of monsoon372

rainfall prevents a reasonable representation of the spring maximum; a rainy season which373

lasts too long can have a similar impact on the fall maximum. The magnitude of the summer374

cooling is also diversely simulated. The spread among AMIP simulations appears to be large375

all year long and particular pronounced outside of the rainy summer months: it reaches up376

to 7 K in winter between the colder and the warmer simulations. As mentionned above, the377

disagreement among models is larger in this semi-arid region that in the wetter soudanian378

region. The spread in temperature is also generally smaller when a low-atmospheric layer379

is considered instead of the 2-m level (not shown). These features suggest that surface380

turbulent and radiative processes play a role in this result, especially during nighttime.381

5. Towards a physical evaluation of the WAM in AMIP382

simulations383

The previous section mainly addressed general and large-scale features of the west African384

monsoon. Higher-frequency fluctuations and finer-scale processes are now evaluated. These385

15

scales are also crucial to improve food management and disaster mitigation in the region386

(e.g., Sultan et al. 2005), as well as the evolution of their characteristics in the climate387

change perspective is key for adaptation policies. Their representation by state-of-the-art388

climate models with the appropriate realism should be one of the key targets if one wants389

them to be trustworthy for simulating either present-day climate or the impact of global390

warming over West Africa.391

a. Intraseasonal variability of precipitation392

Rainfall over West Africa is highly intermittent in space and time. The rainy season is393

punctuated by dry and wet periods occurring at various intraseasonal timescales (Janicot394

et al. 2011). Three preferred timescales have been highlighted: around 40 days, probably395

involving the Madden-Julian Oscillation (Mathews 2004, Janicot et al. 2009), approximately396

15 days with two main regional modes (Mounier and Janicot 2004; Mounier et al. 2008;397

Janicot et al. 2010; Roehrig et al. 2011), and in the 3-10-day range with the well-known398

African Easterly Waves (e.g., Kiladis et al. 2006). In the present study, we do not address399

specifically each of these intraseasonal scales. In contrast, we give a brief overview of the400

main properties of convection intraseasonal variability, which, from this perspective, makes401

West Africa a unique place in the world.402

Figure 11 indicates the variance of OLR filtered in the 1-90-day range. OLR is preferred403

to precipitation for this diagnostic as precipitation variance is closely related to its mean404

value, so that differences in precipitation variance in models are mainly attributed to bias405

in precipitation mean state. A zonally-elongated maximum of OLR variance (> 1000 W2406

m−4) is observed over the Sahel, along the northern side of the ITCZ. When reaching the407

Atlantic ocean, the band moves southward, up to 10◦N. Intraseasonal variance is slightly408

weaker (900 W2 m−4) over the eastern Sahel and Central Africa. Very few models capture409

the structure and amplitude over the Sahel. GFDL-HIRAM-C180 and C360, MRI-CGCM3410

and NorESM1-M overestimate the amount of intraseasonal variability, with a maximum411

16

rather collocated within the ITCZ. The southward slope in the east-west direction is gen-412

erally properly reproduced. However, more than one half of CMIP5 models underestimate413

convective intraseasonal variability. Some of them have a variance reaching one third of that414

observed, when averaged over the domain [5N-20,10W-10E] (Fig. 12.a). These results are415

very similar to those obtained with the CMIP3 models (Roehrig 2010).416

The complete observed distribution of the OLR intraseasonal variance is captured by none417

of the models (Fig. 12.b). The 10-90-day scale explains 20% of the intraseasonal variability418

in only four models (GFDL-HIRAM-C180 and C360, FGOALS-g2 and MIROC5), while it419

is overestimated by more than 10% in the others. Overall, models with underestimation420

of OLR intraseasonal variance put too much weight in long timescales, with the noticeable421

exception of NorESM1-M. The 3-10-day synoptic timescale corresponds to about 50% of the422

observed intraseasonal variance. All models reproduce this amount at an accuracy of ±10%.423

However, the very high-frequency (< 3 days) proportion of almost 30% is not simulated by424

any of the models. Even the very high-resolutions of GFDL-HIRAM-C180 and C360 reach425

only 20%. A high variability at scales < 3 days indicates highly intermittent precipitation426

and very little persistence of them over the Sahel.427

This notion can be quantitatively characterized by the autocorrelation function of pre-428

cipitation. Using it, Lin et al. (2006) showed that the Madden-Julian Oscillation variance in429

most of CMIP3 models comes from an overreddened spectrum, associated with too strong430

persistence of equatorial precipitation. In that regard, Africa has relatively unique properties431

of rainfall. There, precipitation at the grid-point scale are very similar to a white noise, with432

an 1-day lag autocorrelation even negative in some places (Fig. 13.a). There is no persis-433

tence at all at the local scale. No region around the world behaves similarly, either in boreal434

summer (Fig. 13.a) or in boreal winter (Fig. 13.b), except to some extent the northern part435

of Amazonia. Fig. 13.c confirms that such a property of the west African monsoon remains a436

challenge for most of state-of-the-art models. GFDL-HIRAM-C360 reaches the closest value437

to zero ( 0.06), and is closely followed by GFDL-HIRAM-C180 ( 0.09), FGOALS-g2 ( 0.09)438

17

and MIROC5 ( 0.12).439

The success of the GFDL-HIRAMmodel might be attributed to their very high resolution.440

The correct behavior of MIROC5 is likely related to the effort undertaken to make the441

convective scheme more sensitive to dry air in the free troposphere (Chikira and Sugiyama442

2010), and which eliminates the artificial triggering function7 for deep convection used in443

the previous version (CMIP3) of the MIROC model and based on the work of Emori et al.444

(2001). In CMIP3, miroc3.2(medres) and miroc3.2(hires) had a similar behavior to MIROC5445

with regards to this diagnostic (Roehrig 2010). The CMIP3 MRI model (mri_cgcm2.3.2a)446

which was sharing the same convective parameterization (Pan and Randall 1998), except for447

this artificial triggering, produced too much persistence of precipitation over West Africa.448

b. The diurnal cycle at selected AMMA sites449

The CMIP5 archive contains for a few AMIP simulations a large set of diagnostics at450

high-temporal frequency, for ten grid-points along the west African transect. Such high-451

frequency output are useful to evaluate fine-scale processes, and to attempt to relate their452

representation to the large-scale simulation of the monsoon.453

1) Precipitation454

Rain over West Africa is mainly of convective origin (Mathon et al. 2002). Convective455

system properties (size, lifecycle, organization) strongly depend on latitude leading to differ-456

ent characterstics of the diurnal cycle of precipitation. Figure 14 illustrates such differences457

between two sites distant from less than 500 km: the Ouémé site (9.5◦N), just along the458

southern fringe of the ITCZ, and the Niamey site (13.5◦N). The southernmost site presents459

a bimodal diurnal cycle with a first peak around 1800 UTC and a smaller one between 0300460

7The triggering of deep convection occurred only when the relative humidity averaged over the vertical

went over a given threshold ( 80%)

18

and 0600 UTC. Further north, only the morning peak remains, with some precipitation oc-461

curring in the late afternoon. Over Niamey, about of 80% of the annual rainfall is produced462

by westward propagating systems (Dhonneur 1981), born in the afternoon over the elevated463

terrain in northeastern Niger and arriving in the Niamey region in the early morning (Rick-464

enbach et al. 2009). South of the ITCZ, the contribution of propagating systems decreases465

to 50% (Fink et al. 2002; Depraetere et al. 2009), so that a more typical peak in the late466

afternoon can arise. This bimodal structure is also consistent with the secondary nighttime467

peak emerging from global datasets (Yang et al. 2008).468

To our knowledge, no model explicitly includes a proper representation of propagating469

convective systems so that there is no expectation that they can capture the diurnal cycle of470

precipitation over Niamey. Therefore, the use of the Ouémé site as a reference appears more471

suitable for evaluating CMIP5 model diurnal cycle over West Africa (Fig. 14). Across the472

models, rainfall distribution over the Ouémé site is very similar to that over Niamey, when473

it is rescaled by the mean rainfall amount (not shown).474

The distribution of precipitation in CMIP5 models peaks in afternoon. A first group475

of models (CNRM-CM5, HadGEM2-A and MPI-ESM-LR) present a usual too early peak476

in precipitation, between 1200 and 1500 UTC, in phase with insolation and consistently477

with previous studies (Betts and Jakob 2002; Guichard et al. 2004). Nikulin et al. (2012)478

also show that the regional climate models of the CORDEX-Africa experiments do not479

capture the phase of the diurnal cycle. This incorrect timing of the convection maximum480

has consequences on the water cycle at the surface, since larger evaporation will occur with481

a noon rainfall maximum, than with one in the late afternoon (Del Genio 2011).482

The two remaining models (IPSL-CM5B-LR and MRI-CGCM3) exhibits some hope for483

this long-standing issue of climate models. Their maximum is later, between 1500 and 1800484

UTC. Rio et al. (2012) and Sane et al. (2012) already documented this new behavior of485

IPSL-CM5B-LR, which was attributed to a more realistic description of thermal plumes in486

the boundary layer (Rio and Hourdin 2008) and cold pools (Grandpeix and Lafore 2010) as487

19

well as a modified closure and triggering for convection (Rio et al. 2012).488

There is also a large spread in the amplitude and intensity of this afternoon maximum,489

hence affecting the distribution of rain intensity (Fig. 14). Most models present a maximum490

frequency of occurrence for an intensity near 1 mm h−1, reaching even larger values in491

MPI-ESM-LR. In the remaining part of the diurnal cycle, rain intensity decreases by one492

to two orders of magnitude in IPSL-CM5B-LR, HadGEM2-A or MPI-ESM-LR, and by less493

than one in CNRM-CM5 or MRI-CGCM3. These two models have a substantial amount of494

precipitation most of the day. In particular, precipitation intensity lower than 5.10−6 mm495

h−1, which corresponds to the no-rain events in models and observations occurs less than 4%496

in CNRM-CM5 and 26% in MRI-CGCM3, whereas observations display a 88% frequency of497

occurrence.498

2) Fine-scale properties of cloud cover499

The diurnal cycle of cloud covers part of the water cycle, but also of the surface energy500

balance, as it impact the surface incoming shortwave flux in particular. Bouniol et al.501

(2012) highlighted that each of the four cloud categories occurring over the Sahel presents502

a well-marked diurnal cycle (Fig. 15). Two peaks of convective cloud occurrence can be503

identified (about 0900 and 1500 UTC), consistently with the arrival of propagating convective504

systems and locally-initiated convection. Low-level clouds increase between 0900 and 1600505

UT, following the growth of the boundary layer. The maximum in mid-level cloud cover506

occurs between 0300 and 0600 UTC. Cirrus cloud cover decreases between 1200 and 1500507

UTC. The distribution of cloud fraction, also displayed on Fig. 15, highlights the distinct508

cloud fractions associated with each cloud type: low-level and cirrus clouds are relatively509

broken, while mid-level and convective clouds are associated with high cloud fractions.510

Consistently with the diurnal cycle of precipitation, clouds associated with convection511

are shifted towards midday in particular in CNRM-CM5,HadGEM2-A and MPI-ESM-LR,512

where it results in a too early minimum high-level cloud cover. It is followed by a strong513

20

occurrence of deep clouds resulting from condensates detrained from convective updraft and514

treated in most models as a stratiform cloud with no dynamics (Del Genio 2011).515

IPSL-CM5B-LR and MRI-CGCM3 shows to some extent a mid-level cloud occurrence516

with the right timing, but with inaccurate frequencies of occurrence. The diurnal cycle of517

low-level clouds is properly represented in CNRM-CM5 even if the growth of the boundary518

layer seems to be slightly underestimated. HadGEM2-A and MRI-CGCM3 have a very low519

occurrence of these clouds, much too early. MPI-ESM-LR misses this kind of clouds. IPSL-520

CM5B-LR also misses them at Niamey, but due to its too south ITCZ, captures them more521

to the south.522

The statistics in cloud fraction associated with the various cloud types are very different523

from one model to another. IPSL-CM5B-LR and HadGEM2 simulates only broken clouds for524

deep and mid-level clouds, whereas MPI-ESM-LR presents only high cloud fraction values for525

shallow clouds. CNRM-CM5 has a bimodal distribution, with however a frequent unphysical526

65% value for the deep cloud fraction.527

3) Incoming radiation at the surface528

Incoming radiative fluxes at the surface are important components of the surface energy529

budget, with the advantage that they can be reasonably evaluated with a joint utilization530

of in-situ AMMA measurements and satellite products. Their understanding is complex as531

they undergo the influence of the whole troposphere thermodynamical state, the vertical532

distribution of cloud properties, and the aerosol loading. During the monsoon season, the533

latter is expected to impact less the ITCZ region as aerosols are scavenged by precipitation.534

However they strongly affect the Sahara region (Knippertz and Todd 2012).535

At the surface, the JAS shortwave incoming flux meridional gradient is large, reaching536

more than 100 W m−2 from the Guinean coast to the Sahara, and involving a strong CRE537

(Fig. 16). These variations are not accurately simulated, with departures from observations538

larger than several tens of W m−2. Over the Guinea coast, a large number of models539

21

underestimate the incoming shortwave flux in response to a too thin and reflective cloud540

layer. Within the ITCZ, the IPSL models and HadGEM2-A strongly overestimate this541

radiative flux. These two models display a reasonnable cloud frequency of occurrence, but542

they both systematically underestimate the cloud fraction (Fig. 15) . Over the Sahara, most543

of the models overestimates the incoming shortwave flux. Figure 7 pointed there a clear544

deficit of mid-level cloudiness, which has a strong impact in the shortwave (Bouniol et al.545

2012). The description of aerosols may also explain a large part of the spread.546

Meridional fluctuations of the longwave incoming radiation at the surface are much547

weaker. It increases from the more humid and cloudier Gulf of Guinea to the drier Sa-548

hara at 20◦N by 10-20 W m−2. Note however that the atmosphere warms and loads with549

aerosols along this direction. Several models simulate this weak gradient, with little de-550

parture (< 20 W m−2). Further north, biases in the longwave incoming surface radiation551

increase significantly and reach the same order of magnitude as in the shortwave. The lack of552

mid-level clouds may partly explain the underestimation. Further investigations are needed553

to better understand the origin of the spread.554

Over this relative dry region, the feedback onto surface temperature is investigated in the555

models on Fig. 17. Opposite behaviors are noted with increase (decrease) in the temperature556

as the longwave (shortwave) increases with a higher correlation in the longwave domain.557

Figure 17.c shows that the higher the incoming shortwave the lower the incoming longwave.558

Since most of models are simulating a cloud free troposphere in a relative dry environment,559

this suggests a important role of the aerosol loading and their interaction with radiation560

(Kothe and Ahrens 2010) in spread of models over the Sahara. Radiative transfer calculations561

would help in better identifying the sources of discrepancies between models over the region.562

22

6. Summary and conclusions563

In this paper, we present a comprehensive analysis of the representation of the West564

African monsoon in the recently available CMIP5 simulations of both present-day and fu-565

ture climates. The behavior of the models over the Sahel region is examined across a range of566

timescales, going from climate change projections, multi-decadal and interannual variability567

to the intraseasonal and diurnal fluctuations. A specific emphasis is made on the compre-568

hensive set of observational data now available (in particular AMMA and satellite data) to569

evaluate in depth and improve the WAM representation across those scales.570

CMIP5 climate change projections in surface air temperature and precipitation show high571

similarities with those of CMIP3. A robust tendency to warming over the Sahel is noticed572

(about 4 K on average in the RCP8.5 scenario), larger by 10 to 50% compared to the global573

warming. As in CMIP3, the spread of model projections remains large, and models still do574

not agree on the sign of precipitation changes. Most models simulates a moderate rainfall575

response (±20%), lower than the observed decadal variability in the twentieth century. The576

consequences of changes in temperature might therefore be more dramatic. Three outliers577

are highlighted, predicting a large rainfall increase greater than 70%, which tends to cancel578

part of the Sahel warming during the summer monsoon. In contrast, two CMIP3 models579

predict a relatively strong drying of the Sahel, around 40%. Further investigation on the580

rainfall response mechanisms in those models should help to assess their credibility.581

CMIP5 coupled models still suffer of major SST biases in the equatorial Atlantic, which582

induce a systematic southward shift of the ITCZ during the summer, in most models, when583

they are compared to their AMIP counterpart. The similarity between these biases in CMIP3584

and CMIP5 questions the definition of the priorities in climate modeling reasearch programs.585

The ability of coupled models to simulate the multi-decadal and interannual variability586

is assessed in AMIP, historical and pre-industrial control runs. Decadal variability of the587

twentieth century is underestimated in most of the last two types of experiments. In AMIP588

simulations, most models succeed in capturing the partial recovery of monsoon rainfall of the589

23

recent decades, consistently with the role of SSTs in forcing Sahel precipitation (Giannini590

et al. 2003; Biasutti et al. 2008), as well as the amplitude of year-to-year fluctuations. The591

AMIP time sequence is however too short to get rid of internal variability, and ensemble592

AMIP simulations should be useful for further analysis.593

Because of these strong biases in coupled versions, further evaluation of the WAM is per-594

formed in SST-imposed CMIP5 simulations. Almost all of them capture the broad features595

of a monsoon, with various degree of accuracy, depending on the scale that is addressed:596

• The averaged Sahel rainfall exhibits a large spread (±50%).597

• The dispersion in surface air temperature is large too in the Sahel and over the Sahara598

region, and the representation of the summer Saharan heat low appears as a major599

driver of the monsoon latitudinal position. The representation of the radiative aerosol600

properties and surface albedo in this arid region may explain part of this spread.601

• The meridional structure of cloud cover, and its radiative impact, are tough challenges602

for CMIP5 models. This leads to large biases in the surface energy balance, which are603

likely to feedback on the monsoon at large scale.604

• The annual cycle of precipitation is broadly captured by CMIP5 models, but that of605

temperature exhibits a wide dispersion (greater than 5 K). This points to the impor-606

tance of physical processes in the seasonal dynamics of temperature, and questions607

some conclusions that could be drawn from models about the climate sensitivity of the608

phase and amplitude of the temperature annual cycle over the Sahel.609

• The intermittence of precipitation over West Africa is large and only a few models re-610

produce it and more broadly the main features of intraseasonal variability of convection611

there. Results from the GFDL-HIRAM models suggest that intraseasonal variability612

is improved with higher resolution but not necessary the WAM mean state. The rel-613

atively good behavior of a few models in terms of intermittence of precipitation over614

24

the region seems to indicate that models should properly handle the sensitivity of615

convection to humidity in the free troposphere.616

The fine-scale properties of rainfall and clouds are further evaluated at selected sites along617

the AMMA meridional transect, for which high-frequency physical diagnostics were provided618

by some CMIP5 models. It appears that the phasing of diurnal cycle of precipitation with619

the maximum insolation remains a major issue, even though some radical improvements can620

be noticed in two models. However, most of precipitation over the Sahel is provided by621

large mesoscale propagating systems, whose representation is still a challenge for climate622

modelers.623

To summarize, even though most of CMIP5 models capture many features of the west624

African monsoon, they have not reached yet a degree of maturity which makes it possible625

to rely directly on them to anticipate climate changes and their impacts, especially with626

regards to rainfall. Systematic and robust biases of the coupled and atmospheric models have627

not evolved from CMIP3 to CMIP5 and this weakens our confidence in climate projection628

over West Africa. Initiatives which aim at addressing them should be encouraged by the629

community. Some hope regarding the diurnal cycle of convection should be emphasized,630

however, as shown by the results of the IPSL-CM modeling team.631

Finally, the Saharan heat low is also a major actor of the WAM. The spread in its632

representation is large among CMIP5 models, but also among the various observational or633

reanalyzed datasets. This latter point precludes any detailed evaluation for models over634

the Sahara. Experiments such as the recent Fennec field campaign, which provide new635

observations over the Sahara region (Washington et al. 2012), as well as more advanced636

utilization of in-situ and satellite datasets should help advance on this side.637

25

Acknowledgments.638

Based on a French initiative, AMMA was built by an international scientific group and639

is currently funded by a large number of agencies, especially from France, UK, US and640

Africa, and by an EU program. Detailed information on scientific coordination and funding641

is available on the AMMA International web site http://www.amma-international.org.642

Florence Favot and Sophie Tyteca are greatly acknowledged for their support.643

We wish particularly to thank the Principal Investigators who make the various ground-644

based measurements data sets available to the AMMA data base: Sylvie Galle, Colin LLoyd,645

Bernard Cappelaere, Franck Timouk and Laurent Kergoat for the surface radiation measure-646

ments, Théo Vischel, Marielle Gosset, Guillaume Quantin and Eric Mougin for the rain gauge647

network measurements.648

The Pirata buoy measurement were provided by TAO Project Office of NOAA/PMEL.649

The Niamey AMF data were obtained from the Atmospheric Radiation Measurement650

(ARM) Program Archive of the US Department of Energy. CloudSat data are obtained651

from CIRA of Colorado State University. ICARE and NASA give access to the CALIOP652

data. The SRB data were obtained from the NASA Langley Research Center Atmospheric653

Sciences Data Center NASA/GEWEX SRB Project. CMAP Precipitation and Interpolated654

OLR data were provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, USA, from655

their Web site at http://www.esrl.noaa.gov/psd/.656

We acknowledge the World Climate Research Programme’s Working Group on Coupled657

Modelling, which is responsible for CMIP, and we thank the climate modeling groups (listed658

in Table 1 of this paper) for producing and making available their model output. For CMIP,659

the U.S. Department of Energy’s Program for Climate Model Diagnosis and Intercomparison660

provides coordinating support and led development of software infrastructure in partnership661

with the Global Organization for Earth System Science Portals.662

Dominique Bouniol and Francoise Guichard were financially supported by CNES and the663

EUCLIPSE program from the European Union, Seventh Framwork Programme (FP7/2007–664

26

2013) un der grant agreement n◦244067.665

27

666

REFERENCES667

Bao, X. and F. Zhang, 2012: Evaluation of NCEP/CFSR, NCEP/NCAR, ERA-Interim and668

ERA-40 reanalysis datasets against independent sounding observations over the Tibetan669

Plateau. J. Climate, doi:10.1175/JCLI-D-12-00056.1.670

Betts, A. and C. Jakob, 2002: Study of diurnal cycle of convective precipitation over671

Amazonia using a single column model. J. Geophys. Res., 107 (D23), D23, doi:672

10.1029/2002JD002264.673

Biasutti, M. and A. Giannini, 2006: Robust Sahel drying in response to late 20th century674

forcings. Geophys. Res. Lett., 33, L11706, doi:10.1029/2006GL026067.675

Biasutti, M., I. M. Held, A. H. Sobel, and A. Giannini, 2008: SST forcings and Sahel676

rainfall variability in simulations of the twentieth and twenty-first centuries. J. Climate,677

21, 3471–3486.678

Bony, S., M. Webb, C. Bretherton, S. Klein, P. Siebesma, G. Tselioudis, and M. Zhang, 2011:679

CFMIP: towards a better evaluation and understanding of clouds and cloud feedbacks in680

CMIP5 models. CLIVAR Exchanges No 56, 16 (2), 20–24.681

Boone, A. A., I. Poccard-Leclercq, Y. Xue, J. Feng, and P. de Rosnay, 2010: Evaluation682

of the WAMME model surface fluxes using results from de AMMA land-surface model683

intercomparison project. Clim. Dyn., 35, 127–142, doi:10.1007/s00382-009-0653-1.684

Bouniol, D., F. Couvreux, P.-H. Kamsu-Tamo, M. Leplay, F. Guichard, F. Favot, and685

E. O’Connor, 2012: Diurnal and seaonal cycles of cloud occurrences, types and ra-686

diative impact over West Africa. J. Appl. Meteor. Climatol., 51, 534–553, doi:10.1175/687

JAMC-D-11-051.1.688

28

Bouniol, D., J. Delanoë, C. Duroure, A. Protat, V. Giraud, and G. Penide, 2010: Microphys-689

ical characterisation of west African MCS anvils. Quart. J. Roy. Meteor. Soc., 136 (s1),690

323–344, doi:10.1002/qj557.691

Chauvin, F., R. Roehrig, and J.-P. Lafore, 2010: Intraseasonal variability of the Saharan692

Heat Low and its link with mid-latitudes. J. Climate, 23, 2544–2561.693

Chikira, M. and M. Sugiyama, 2010: A cumulus parameterization with state-dependent694

entrainment rate. Part I: description and sensitivity to temperature and humidity profiles.695

J. Atmos. Sci., 67, 2171–2193.696

Chou, C. and J. D. Neelin, 2003: Mechanisms limiting the northward extent of the nothern697

summer monsoons over North America, Asia, and Africa. J. Climate, 16, 406–425.698

Chou, C., J. D. Neelin, C.-A. Chen, and J.-Y. Tu, 2009: Evaluating the rich-get-recher699

mechanism in tropical precipitation change under global warming. J. Climate, 22, 1982–700

2005.701

Collins, W. J., et al., 2008: Evalution of the HadGEM2 model. Technical note 74, Meteo-702

rological Office Hadley Centre, Exeter, England. URL http://www.metoffice.gov.uk/703

publications/HCTN/HCTN_74.pdf.704

Cook, K. H., 2008: Climate science: The mysteries of Sahel droughts. Nat. Geosci., 1.,705

647–648.706

Cook, K. H. and E. K. Vizy, 2006: Coupled model simulations of the West African monsoon707

system: twentieth- and twenty-first-century simulations. J. Climate, 19, 3681–3703.708

Del Genio, A. D., 2011: Representing the sensitivity of convective cloud systems to tro-709

pospheric humidity in general circulation models. Surveys in Geophysics, doi:10.1007/710

s10712-011-9148-9.711

29

Depraetere, C., M. Gosse, S. Ploix, and H. Laurent, 2009: The organization and kinematics712

of tropical rainfall systems gournd tracked at mesoscale with gages: first results from the713

campaign 1999–2006 on the Upper Ouémé Valley (Benin). J. Hydrol., 375 (1-2), 143–160,714

doi:10.1016/j.hydrol.2009.01.011.715

Dhonneur, G., 1981: Les amas nuageux mobiles: principale composante de la métérologie716

du Sahel. La Météorologie, 27, 75–82.717

Donner, L. J., et al., 2011: The dynamical core, physical paramterizations, and basic simu-718

lation characteristics of the atmospheric component of the GFDL glocabl coupled model719

CM3. J. Climate, 24, 3484–3518.720

Dufresne, J.-L., et al., 2012: Climate change projections using the IPSL-CM5 Earth system721

model: from CMIP3 to CMIP5. Clim. Dyn.722

Emori, S., T. Nozawa, A. Numaguti, and I. Uno, 2001: Importance of cumulus parameter-723

ization for precipitation simulation over East Asia in June. J. Meteor. Soc. Japan, 79,724

939–947.725

Fan, Y. and H. van den Dool, 2008: A global monthly land surface air temperatire analysis726

for 1948-present. J. Geophys. Res., 113, D1103, doi:10.1029/2007JD008470.727

Fink, A. and A. Reiner, 2003: Spatio-temporal variability of the relation between African728

easterly waves and West African squall lines in 1998–1999. J. Geophys. Res., 108 (D11),729

4332, doi:10.1029/2002JD002816.730

Fink, A. H., D. G. Vincent, and V. Emert, 2002: Rainfall types in the West African Sudanian731

zone during the summer monsoon 2002. Mon. Wea. Rev., 134, 2143–2164, doi:10.1175/732

MWR3182.1.733

Gent, P. R., et al., 2011: The Community Climate System Model Version 4. J. Climate,734

24 (19), 4973–4991.735

30

Giannini, A., R. Saravanan, and P. Chang, 2003: Oceanic forcing of Sahel rainfall on inter-736

annual to interdecadal timescales. Science, 302, 1027–1030.737

Giorgetta, M. A., et al., 2012: The atmospheric general circulation model ECHAM6: model738

description. Tech. rep., Max Plank Institut für Meteorologie, Hamburg, Germany. URL739

http://www.mpimet.mpg.de/en/science/models/echam.html.740

Grandpeix, J.-Y. and J.-P. Lafore, 2010: A density current parameterization coupled with741

Emanuel’s convection scheme. Part I: the models. J. Atmos. Sci., 67 (4), 881–897.742

Guichard, F., L. Kergoat, E. Mougin, F. Timouk, F. Baup, P. Hiernaux, and F. Lavenu,743

2009: Surface thermodynamics and radiative budget in the Sahelian Gourma: seasonal744

and diurnal cycles. J. Hydrol., 375 (1-2), 161–177.745

Guichard, F. et al., 2004: Modeling the diurnal cycle of deep precipitating convection over746

land with cloud-resolving models and single-column models. Quart. J. Roy. Meteor. Soc.,747

130, 3139–3172.748

Hazeleger, W., et al., 2010: EC-Earth: a seamless Earth system prediction approach in749

action. Bull. Amer. Meteor. Soc., 91, 1357–1363, doi:10.1175/2010BAMS2877.1.750

Held, I. M., T. L. Delworth, J. Lu, K. L. Findell, and T. R. Knutson, 2005: Simulation751

of Sahel drought in the 20th and 21th centuries. Proc. Natl. Acad. Sci. U. S. A., 102,752

17 891–17 896.753

Hoerling, M., J. Hurrell, J. Eischeid, and A. Phillips, 2006: Detection and attribution of754

twentieth-century northern and southern African rainfall change. J. Climate, 19, 3989–755

4008.756

Hourdin, F., et al., 2010: AMMA-Model Intercomparison Project. Bull. Amer. Meteor. Soc.,757

91 (1), 95–104.758

31

Hourdin, F., et al., 2012: LMDZ5B: the atmospheric component of the IPSL climate model759

with revisited parameterizations for clouds and convection. Clim. Dyn.760

Huffman, G. J., R. F. Adler, M. M. Morrissey, D. T. Bolvin, S. Curtis, R. Joyce, B. Mc-761

Gavock, and J. Susskind, 2001: Global precipitation of one-degree daily resolution from762

multisatellite observations. J. Hydrometeor., 2, 36–50.763

Huffman, G. J., D. T. Bolvin, E. J. Nelkin, D. B. Wolff, G. Gu, Y. Hong, K. P. Bowman, and764

E. F. Stocker, 2007: The TRMM Multisatellite Precipitation Analysis (TMPA): Quasi-765

global, multiyear, combined-sensor precipitation estimates at fine scales. J. Hydrometeor.,766

8 (1), 38–55.767

Janicot, S., F. Mounier, S. Gervois, B. Sultan, and G. Kiladis, 2010: The dynamics of768

the West African monsoon. Part V: The detection and role of the dominant modes of769

convectively coupled equatorial Rossby waves. J. Climate, 23, 4005–4024.770

Janicot, S., et al., 2011: Intraseasonal variability of the West African monsoon. Atmos. Sci.771

Let., 12 (1), 58–66, doi:10.1002/asl.280.772

Jones, C. G., F. Giorgi, and G. Asrar, 2011: The Coordinated Regional Downscaling Exper-773

iment: CORDEX. An international downscaling link to CMIP5. CLIVAR Exchanges No774

56, 16 (2), 34–400.775

Kiladis, G. N., C. D. Thorncroft, and N. M. J. Hall, 2006: Three-dimensional structure and776

dynamics of African easterly waves. Part I: Obsevations. J. Atmos. Sci., 63, 2212–2230.777

Kirkevåg, A., T. Iversen, O. Seland, J. B. Debernard, T. Storelvmo, and J. E. Kristjánsson,778

2008: Aerosol-cloud-climate interactions in the climate model CAM-Oslo. Tellus, 60A,779

492–512.780

Knippertz, P. and M. C. Todd, 2012: Mineral dust aerosols over the Sahara: meteorological781

32

controls on emission and transport and implications for modeling. Rev. Geophys., 50,782

RG1007, doi:10.1029/2011RG000362.783

Kothe, S. and B. Ahrens, 2010: On the radiation budget in regional climate simulations of784

West Africa. J. Geophys. Res., 115, D23120, doi:10.1029/2010JD014331.785

Lau, K. M., S. S. P. Shen, K.-M. Kim, and H. Wang, 2006: A multimodel study of the786

twentieth-century simulations of Sahel drought from the 1970s to 1990s. J. Geophys. Res.,787

111, D07111, doi:10.1029/2005JD006281.788

Lavaysse, C., C. Flamant, S. Janicot, D. J. Parker, J.-P. Lafore, B. Sultan, and J. Pelon,789

2009: Seasonal evolution of the West African heat low: a climatological perspective. Clim.790

Dyn., 33 (2-3), 313–330, doi:10.1007/s00382-009-0553-4.791

Le Lay, M. and S. Galle, 2005: Seasonal cycle and interannual variability of rainfall at hy-792

drological scales. The West African monsoon in a Sudanese climate. Hydrological Sciences793

Journal, 50, 3509–3524.794

Lebel, T., et al., 2010: The AMMA field campaigns: Multiscale and multidisciplinary obser-795

vations in the West African region. Quart. J. Roy. Meteor. Soc., 136 (s1), 8–33.796

Liebmann, B. and C. A. Smith, 1996: Description of a complete (interpolated) outgoing797

longwave radiation dataset. Bull. Amer. Meteor. Soc., 77, 1275–1277.798

Lin, J.-L. et al., 2006: Tropical intraseasonal variability in 14 IPCC AR4 climate models.799

Part I: Convective signals. J. Climate, 19, 2665–2690.800

Loeb, N. G., B. A. Wielicki, D. R. Doelling, G. L. Smith, D. F. Keyes, S. Kato, N. Manalo-801

Smith, and T. Wong, 2009: Towards optimal closure of the Earth’s top-of-atmosphere802

radiation budget. J. Climate, 22, 748–766, doi:10.1175/2008JCLI2637.1.803

Mathon, V., H. Laurent, and T. Lebel, 2002: Mesoscale convective system rainfall in the804

Sahel. J. Appl. Meteorol., 41, 1081–1092.805

33

Meehl, G. and Coauthors, 2007: Global Climate Projection. Climate Change 2007: The806

Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Re-807

port of the Intergovernmental Panel on Climate Change, S. Solomon, D. Qin, M. Manning,808

Z. Chen, M. Marquis, K. B. Averyt, M. Tignor, and H. L. Miller, Eds., Cambridge Uni-809

versity Press, Cambridge, United Kingdom and New York, NY, USA, 747–846.810

Meehl, G. A., C. Covey, T. Delworth, M. Latif, B. McAvaney, J. F. B. Mitchell, R. J.811

Stouffer, and K. E. Taylor, 2007: The WCRP CMIP3 multimodel dataset: A new era in812

climate change research. Bull. Amer. Meteor. Soc., 88, 1383–1394.813

Miller, M. A. and A. Slingo, 2007: The ARM Mobile Facility and its first international814

deployment: measuring flux divergence in West Africa. Bull. Amer. Meteor. Soc., 88,815

1229–1244, doi:10.1175/BAMS-88-8-1229.816

Mitchell, T. D. and P. D. Jones, 2005: An improved method of constructing a database817

of monthly climate observations and associated high-resolution grids. Int. J. Climatol.,818

25 (6), 673–712, doi:10.1002/joc.1181.819

Mizuta, R., et al., 2012: Climate simulations using MRI-AGCM3.2 with 20-km grid. J.820

Meteor. Soc. Japan, 90A, 233–258.821

Mougin, E., et al., 2009: The AMMA-CATCH Gourma observatory site in Mali: relating cli-822

matic variations to changees in vegetation, surface hydrology, fluxes and natural resources.823

J. Hydrol., 375, 14–33.824

Mounier, F. and S. Janicot, 2004: Evidence of two independant modes of convection at825

intraseasonal timescale in the West African summer monson. Geophys. Res. Lett., 31,826

L16116, doi:10.1029/2004GL020665.827

Mounier, F., S. Janicot, and G. N. Kiladis, 2008: The West African monsoon dynamics.828

Part III: The quasi-biweekly zonal dipole. J. Climate, 21, 1911–1928.829

34

Neale, R. B., et al., 2010a: Description of the NCAR Community Atmospheric Model830

(CAM4.0). Technical note NCAR/TN-485+STR, National Center for Atmospheric Re-831

search, Boulder, Colorado, USA. URL http://www.cesm.ucar.edu/models/ccsm4.0/832

cam/docs/description/cam4_desc.pdf.833

Neale, R. B., et al., 2010b: Description of the NCAR Community Atmospheric Model834

(CAM5.0). Technical note NCAR/TN-486+STR, National Center for Atmospheric Re-835

search, Boulder, Colorado, USA.836

Nicholson, S., 1980: The nature of rainfall fluctuations in subtropical West Africa. Mon.837

Wea. Rev., 108, 473–487.838

Nicholson, S. E., B. Some, and B. Kone, 2000: An analysis of recent rainfall conditions in839

West Africa, including the rainy seasons if the 1997 El Niño and the 1998 La Niña years.840

J. Climate, 13 (14), 2628–2640.841

Nikulin, G., et al., 2012: Precipitation climatology in an esemble of CORDEX-Africa regional842

climate simulations. J. Climate.843

Paeth, H. and A. Hense, 2004: SST versus climate change signals in West African rainfall:844

20th-century variations and future projections. Climatic Change, 65, 179–208.845

Pan, D.-M. and D. A. Randall, 1998: A cumulus parameterization with prognostic closure.846

Quart. J. Roy. Meteor. Soc., 124, 949–981.847

Redelsperger, J.-L., C. D. Thorncroft, A. Diedhiou, T. Lebel, D. J. Parker, and J. Polcher,848

2006: African Monsoon Multidisciplinary Analysis: an international research project and849

field campaign. Bull. Amer. Meteor. Soc., 87, 1739–1746.850

Richter, I. and S.-P. Xie, 2008: On the origin of equatorial Atlantic biases in coupled general851

circulation models. Clim. Dyn., 31 (5), 587–598, doi:10.1007/s00382-008-0364-z.852

35

Rickenbach, T., R. N. Ferreira, N. Guy, and E. Williams, 2009: Radar-observed squall line853

propagation and the diurnal cycle of convection in Niamey, Niger, during the 2006 African854

Monsoon and Multidisciplinary Analyses Intensive Observing Period. J. Geophys. Res.,855

114, D03107, doi:10.1029/2008JD010871.856

Rienecker, M. M., et al., 2011: MERRA - NASA’s Modern-Era Retrospective Analysis for857

Research and Applications. J. Climate, 24, 3624–3648, doi:10.1175/JCLI-D-11-00015.1.858

Rio, C. and F. Hourdin, 2008: A thermal plume model for convective boundary layer:859

representation of cumulus clouds. J. Atmos. Sci., 65, 407–425.860

Rio, C., et al., 2012: Control of deep convection by sub-cloud lifting processes: the ALP861

closure in the LMDZ5B general circulation model. Clim. Dyn.862

Roehrig, R., 2010: Variabilité intrasainnière de la mousson africaine : Caractérisation et863

Modélisation (Intraseasonal variability of the West African monsoon : Characterization864

and modeling). Ph.D. thesis, Paris-Est University.865

Roehrig, R., F. Chauvin, and J.-P. Lafore, 2011: 10-25-day intraseasonal variability of con-866

vection over the Sahel: a role of the Saharan heat low and midlatitudes. J. Climate, 24,867

5863–5878, doi:10.1175/2011JCLI3960.1.868

Rotstayn, L., M. Collier, M. Dix, Y. Feng, H. Gordon, S. O’Farrell, I. Smith, and J. Syktus,869

2010: Improved simulation of Australian climate and ENSO-related climate variability870

in a GCM with interactive aerosol treatment. Int. J. Climatol, 30 (7), 1067–1088, doi:871

10.1002/joc.1952.872

Ruti, P. M. and A. Dell’Aquila, 2010: The XX century AEJ and AEWs in reanalyses and873

IPCC simulations. Submitted to Climate Dynamics.874

Saha, S., et al., 2010: The NCEP climate forecast system reanalysis. Bull. Amer. Meteor.875

Soc., 91, 1015–1057.876

36

Sane, Y., et al., 2012: Analysis of the diurnal cycle of precipitation over Dakar using local877

rain-gauge data and a general circulation model. Quart. J. Roy. Meteor. Soc., doi:10.1002/878

qj.1932.879

Simmons, A., S. Uppala, D. Dee, and S. Kobayashi, 2007: ERA-Interim: New ECMWF880

reanalysis products from 1989 onwards. ECMWF Newsletter, 110.881

Slingo, A., H. E. White, N. A. Bharmal, and G. J. Robinson, 2009: Overview of observations882

from the RADAGAST experiment in Niamey, Niger: 2. radiative fluxes and divergences.883

J. Geophys. Res., 114, D00E04, doi:10.1029/2008JD010497.884

Slingo, J., 1980: A cloud parameterization scheme derived from GATE data for use with a885

numerical model. Quart. J. Roy. Meteor. Soc., 106, 747–770.886

Stackhouse, P. W., Jr., S. K. Gupta, S. J. Cox, J. C. Mikovitz, T. Zhang, and L. M.887

Hinkelman, 2011: The NASA/GEWEX Surface Radiation Budget Release 3.0: 24.5-year888

dataset. GEWEX News, 21 (1), 10–12.889

Stephens, G. L., et al., 2002: The CloudSat mission and the A-Train: A new dimension890

of space-based observations of clouds and precipitation. Bull. Amer. Meteor. Soc., 83,891

1771–1790.892

Sultan, B., C. Baron, M. Dingkuhn, and S. Janicot, 2005: Agricultural impacts of large-893

scale variability of the West African monsoon. Agricultural and Forest Meteorology, 128,894

93–110.895

Sultan, B. and S. Janicot, 2000: Abrupt shift of the ITCZ over West Africa and intra-seasonal896

variability. Geophys. Res. Lett., 27, 3353–3356.897

Sultan, B. and S. Janicot, 2003: The West African monsoon dynamics. Part II: The “preon-898

set” and “onset” of the summer monsoon. J. Climate, 16, 3407–3427.899

37

Taylor, K. E., R. J. Stouffer, and G. A. Meehl, 2012: An overview of CMIP5 and the experi-900

ment design. Bull. Amer. Meteor. Soc., 93 (4), 485–498, doi:10.1175/BAMS-D-11-00094.901

1.902

Thorncroft, C. D. and M. Blackburn, 1999: Maintenance of the African easterly jet. Quart.903

J. Roy. Meteor. Soc., 125, 763–786.904

Thorncroft, C. D. and D. P. Rowell, 1998: Interannual variability of African wave activity905

in a general circulation model. Int. J. Climatol., 18, 1305–1323.906

Timouk, F., L. Kergoat, E. Mougin, C. Lloyd, E. Ceschia, P. de Rosnay, P. Hiernaux,907

and V. Demarez, 2009: Response of sensible heat flux to water regime and vegetation908

development in a central Sahelian landscape. J. Hydrol., 375, 178–189.909

Tippett, M. and A. Giannini, 2006: Potentially predictable components of African summer910

rainfall in a SST-forced GCM simulation. J. Climate, 19, 3133–3144.911

Voldoire, A., et al., 2012: The CNRM-CM5.1 global climate model: description and basic912

evaluation. Clim. Dyn., doi:10.1007/s00382-011-1259-y.913

Volodin, E. M., N. A. Diansky, and A. V. Gusev, 2010: Simulating present-day climate914

with the INMCM4.0 coupled model of the atmospheric and oceanic general circulations.915

Atmospheric and Oceani Physics, 46, 414–431.916

Waliser, D. E., J.-L. Li, T. L’Ecuyer, and W.-T. Chen, 2011: The impact of precipitating917

ice and snow on the radiation balance in global climate models. Geophys. Res. Lett., 38,918

L06802, doi:10.1029/2010GL046478.919

Wang, W., P. Xie, S.-H. Yoo, Y. Xue, A. Kumar, and X. Wu, 2011: An asseesment of920

the surface climate in the NCEP climate forecast system reanalysis. Clim. Dyn., 37 (7),921

1601–1620, doi:10.1007/s00382-010-0935-7.922

38

Washington, R., et al., 2012: Fennec: the Saharan climate system. 4th international AMMA923

Conference, 2-6 July 2012, Toulouse, France.924

Watanabe, M., et al., 2010: Improved climate simulation by MIROC5: mean states, vari-925

ability, and climate sensitivity. J. Climate, 23, 6312–6335, doi:10.1175/2010JCLI3679.1.926

Winker, D., W. Hunt, and M. McGill, 2007: Initial performance assessment of CALIOP.927

Geophys. Res. Lett., 34, L19803, doi:10.1029/2007GL030135.928

Wu, T., et al., 2010: The Beijing Climate Center for Atmospheric General Circulation929

Model (BCC-AGCM2.0.1: description and its performance for the present-day climate.930

Clim. Dyn., 34, 123–147.931

Xie, P. and P. A. Arkin, 1997: Global precipitation: a 17-year monthly analysis based on932

gauge observation, satellite estimates, and numerical model outputs. Bull. Amer. Meteor.933

Soc., 78, 2539–2558.934

Xue, Y., et al., 2010: Intercomparison and analyses of the climatology of the West African935

Monsoon in the West AFrican Monsoon Modeling and Evaluation project (WAMME) first936

model intercomparison experiment. Clim. Dyn., doi:10.1007/s00382-010-0778-2.937

Yang, S., K.-S. Kuo, and E. A. Smith, 2008: Persistent nature of secondary diurnal modes938

of precipitation over oceanic and continental regimes. J. Climate, 21 (16), 4115–4131.939

Yokimoto, S., et al., 2011: Meteorological Research Institue-Earth System Model v1 (MRI-940

ESM1) - Model Description. Technical report 64, Meteorological Research Institute, Japan.941

39

List of Tables942

1 Available CMIP5 models and simulations. 41943

2 Characterics of AMIP simulation models. 42944

3 Observational datasets. 43945

40

Table 1. Available CMIP5 models and simulations.

Centers Models SimulationsBCC (China) bcc-csm1-1 amip?, hist, rcp4.5, rcp8.5, picontrol

CCCma (Canada)CanAM4 amip?†

CanCM4 histCanESM2 hist, rcp4.5

CMCC (Italy) CMCC-CM hist, rcp4.5CNRM-CERFACS (France) CNRM-CM5 amip?†, hist, rcp8.5, picontrolCSIRO-BOM (Australia) ACCESS1-3 hist, rcp4.5, rcp8.5, picontrolCSIRO-QCCE (Australia) CSIRO-Mk3-6-0 amip?, hist, rcp4.5, rcp8.5, picontrol

FIO (China) FIO-ESM hist, rcp4.5, rcp8.5ICHEC EC-EARTH amip†, hist, rcp4.5, rcp8.5 vÈrifier...

INM (Russia) inmcm4 amip?, hist, rcp4.5, rcp8.5

IPSL (France)IPSL-CM5A-LR amip?†, hist, rcp4.5, rcp8.5, picontrolIPSL-CM5A-MR hist, rcp4.5, rcp8.5IPSL-CM5B-LR amip?†, hist, picontrol

LASG-CESS (China) FGOALS-g2 amip?, hist, rcp4.5, rcp8.5, picontrolLASG-IAP (China) FGOALS-s2 amip?, hist, rcp4.5, picontrol

MIROC (Japan)

MIROC4h histMIROC5 amip?, hist, rcp4.5, rcp8.5, picontrol

MIROC-ESM hist, rcp8.5MIROC-ESM-CHEM hist, rcp4.5, rcp8.5

MOHC (England)

HadCM3 histHadGEM2-A amip?†

HadGEM2-CC hist, rcp8.5HadGEM2-ES hist, rcp4.5, rcp8.5

MPI-M (Germany)MPI-ESM-LR amip?†, hist, rcp4.5, rcp8.5, picontrolMPI-ESM-MR amip, hist, rcp4.5, rcp8.5, picontrolMPI-ESM-P hist

MRI (Japan)MRI-AGCM3-2H amip?

MRI-AGCM3-2S amip?

MRI-CGCM3 amip?†, hist, rcp4.5, rcp8.5, picontrol

NASA-GISS (USA) GISS-E2-H histGISS-E2-R hist, rcp4.5, rcp8.5, picontrol

NCAR (USA) CCSM4 amip?, hist, picontrol

NCC (Norwegia) NorESM1-M amip?, hist, rcp4.5, picontrolNorESM1-ME hist, rcp4.5, rcp8.5

NIMR-KMA (South Korea) HadGEM2-AO hist, rcp4.5

NOAA-GFDL (USA)

GFDL-CM2p1 histGFDL-CM3 hist, rcp4.5

GFDL-ESM2G hist, rcp4.5, rcp8.5GFDL-ESM2M hist, rcp4.5, rcp8.5

GFDL-HIRAM-C180 amip?

GFDL-HIRAM-C360 amip?

NSF-DOE-NCAR (USA)

CESM1-BGC hist, rcp4.5, rcp8.5CESM1-CAM5 amip, hist, rcp4.5, rcp8.5, picontrol

CESM1-FASTCHEM histCESM1-WACCM hist

? Daily outputs are available for these simulations.† High-frequency outputs are used at selected sites.

41

Tabl

e2.

Cha

ractericsof

AMIP

simulationmod

els.

Mod

els

Referen

ceHorizon

tal

Num

berof

Grid

Verticallevel

bcc-csm1-1

Wuet

al.(

2010

)T42

(2.8◦x2

.8◦ )

26Can

AM4

http

://w

ww.e

c.gc

.ca/

ccma

c-cc

cma/

T63

(1.87

5◦x1

.875◦ )

35CNRM-C

M5

Voldo

ireet

al.(20

12)

T12

7(1.4◦x1

.4◦ )

31ACCESS

1-3

http

://w

iki.

csir

o.au

/con

flue

nce/

disp

lay/

ACCE

SS/

1.87

5◦x1

.25◦

38CSIRO-M

k3-6-0

Rotstaynet

al.(

2010)

T63

(1.87

5◦x1

.875◦ )

18EC-E

ARTH

Hazeleger

etal.(20

10)

T15

9(1.12

5◦x1.12

5◦)

62inmcm

4Volod

inet

al.(20

10)

1.5◦x2

.5◦

21IP

SL-C

M5A

-LR

Dufresneet

al.(20

12)

1.89

5◦x3

.75◦

39IP

SL-C

M5B

-LR

Hou

rdin

etal.(20

12)

1.89

5◦x3

.75◦

39FG

OALS

-g2

http

://w

ww.l

asg.

ac.c

n/FG

OALS

/CMI

P52.8◦x2

.8◦

26FG

OALS

-s2

http

://w

ww.l

asg.

ac.c

n/FG

OALS

/CMI

P51.7◦x2

.8◦

26MIR

OC5

Watan

abeet

al.(20

10)

T85

(1.4◦x1

.4◦ )

40Had

GEM2-A

Collin

set

al.(20

08)

1.25◦ x1.87

5◦38

MPI-ESM

-LR

Giorgetta

etal.(20

12)

T63

(1.87

5◦x1

.875◦ )

47MPI-ESM

-MR

Giorgetta

etal.(20

12)

T63

(1.87

5◦x1

.875◦ )

95MRI-AGCM3-2H

Mizutaet

al.(20

12)

T31

9(0.56◦ x0.56◦ )

64MRI-AGCM3-2S

Mizutaet

al.(20

12)

T95

9(0.19◦ x0.19◦

64MRI-CGCM3

Yok

imotoet

al.(20

11)

T15

9(1.12

5◦x1.12

5◦)

48CCSM

4Neale

etal.(20

10a);G

entet

al.(20

11)

0.9◦x1

.25◦

26NorESM

1-M

Kirkevåget

al.(20

08)

1.9◦x2

.5◦

26GFDL-HIR

AM-C

180

Don

neret

al.(20

11)

0.5◦x0

.5◦

32GFDL-HIR

AM-C

360

Don

neret

al.(20

11)

0.25◦ x0.25◦

32CESM

1-CAM5

Neale

etal.(20

10b)

0.5◦x1

.25◦

30

42

Tabl

e3.

Observation

alda

tasets.

Variables

Dataset

References

Resolution

Frequency

Periodcovered

Rainfall

CRU

version3.1

Mitchellan

dJo

nes

(2005)

0.5◦

x0.5◦

mon

thly

1901–2002

CMAP

Xie

andArkin

(1997)

2.5◦

x2.5◦

mon

thly

1979–2008

GPCP

Huffman

etal.(2001)

1◦x1◦

daily

1997–2008

TRMM-3B42

Huffman

etal.(2007)

0.25

◦x0.25

◦daily

1998–2008

Ouém

érain

gauges(2

◦E-9.5

◦N)

LeLay

andGalle

(2005);Depraetereet

al.(2009)

site

a5/30

minutes

1999–2011

Niamey

rain

gauges(2.2

◦E-13.5◦

N)

Leb

elet

al.(2010)

site

a5/30

minutes

1990–2011

Con

vection

NOAA

OLR

Liebman

nan

dSmith(1996)

2.5◦

x2.5◦

daily

1979–2008

Clouds

CloudSat/C

ALIP

SO

Bou

niolet

al.(2012)

AMMA

tran

sect

bmon

thly

2006–2009

ARM

Mob

ileFacility(A

MF)

Milleran

dSlingo

(2007);Bou

niolet

al.(2012)

Niamey

site

30minutes

2006

Rad

iation

CERES-E

BAFcEdition2.6

Loeb

etal.(2009)

1◦x1◦

mon

thly

2000–2010

NASA/G

EWEX

SRB

release3.0(pr/qc)

dStackhou

seet

al.(2011)

1◦x1◦

mon

thly

1983–2007

Transect

sitese

?sites

JASaverage

2-m

tempÈrature

CRU

version3.1

Mitchellan

dJo

nes

(2005)

0.5◦

x0.5◦

mon

thly

1979–2008

NCEP

CFSR

Sah

aet

al.(2010)

1◦x1◦

mon

thly

1979–2008

NASA-M

ERRA

Rieneckeret

al.(2011)

1◦x1◦

mon

thly

1979–2008

ERA-Interim

Sim

mon

set

al.(2007)

0.75

◦x0.75

◦mon

thly

1979–2008

aThenumber

anddensity

ofgaugesvary

from

year

toyear.Tomatch

the30-m

inmodel

outputresolution

sextractedat

theseselected

sites,

andnot

oversample

anyparticularregion

,thedatawere

re-sam

pledan

dspatiallyhom

ogenized

in0.2◦

x0.2◦

dom

ainwithin

2◦x2◦

area

arou

ndbothsites.

bEachsatellitetrackwithin

a10

◦W-10◦

Edom

ainis

assumed

representative

oftheGreenwichlatitude,

lead

ingto

atleasttw

osamplingper

day

alon

gthis

tran

sect.

cTheCloudsan

dEarth’sRad

iantEnergy

System

(CERES)Energy

Balan

cedan

dFilled(E

BAF)Edition2.6product

consistsof

top-of-atmospherean

dsurfaceradiative

fluxes

(htt

p://

cere

s.la

rc.n

asa.

gov).

dTheNASA/G

EWEX

SurfaceRad

iation

Budget(SRB)consistsof

top-of-atmospherean

dsurfaceradiative

fluxes.Twosets

ofsurfaceradiative

fluxestimationsareavailable,based

ondifferentalgorithms,

know

nas

primary(SRB3p

r)an

dLan

gley

param

eterized

algorithms(SRB3q

c).Moreinform

ationon

thesetw

odatasetsareavailable

atht

tp:/

/gew

ex-sr

b.la

rc.n

asa.

gov/

inde

x.ph

p.eBam

ba(17.1◦

N-1.4

◦E,2004–2007),Agoufou(15.34

◦N-1.48◦

E,2002–2008),Wan

kama(13.67

◦N-2.65◦

E,2005–2006),Bira(9.82◦

N-1.71◦

E,2006),

Nah

olou

(9.73◦

N-1.60◦

E,2006)an

dPIR

ATA

Buoy

(0◦-0

◦,2006,on

lyincomingshortw

aveat

surface).

fGlobal

HistoricalClimatologyNetworkversion2an

dClimateAnom

alyMon

itoringSystem

.

43

List of Figures946

1 a) Climate projections of precipitation (in %) plotted against those of 2-meter947

temperature (in K) averaged over a Sahel domain [10◦N-18◦N, 13◦W-10◦E].948

The difference are computed between period 2071–2100 and the period 1971–949

2000, for the JAS season, for CMIP3 SRESA2 scenario and for CMIP5 RCP4.5950

and RCP8.5 scenarios. b) Climate projections of global 2-meter temperature951

plotted against those of 2-meter temperature averaged over the Sahel domain952

[10◦N-18◦N, 13◦W-10◦E]. Red points indicate projections for which precipita-953

tion change is greater than 15%. c) Same as b), but for the MAM season. 48954

2 a) Time evolution of raw (thin line with dots) precipitation in mm day1 av-955

eraged over [10◦N-18◦N, 13◦W-10◦E] and of its decadal component P 9 (thick956

solid line), computed as the 9-year running mean of the raw index. CRU957

data is indicated by the thick black line and CMAP by the thin black line.958

b) Mean difference of P 9 between the 199-2004 and 1983–1988 periods for959

AMIP simulations and CRU precipitations (dots), and standard deviation of960

P9 in pre-industrial control (open squares) and historical (filled squares) ex-961

periments. The standard deviation has been normalized by the mean P 9. For962

observation, the filled red square corresponds to the standard deviation of963

CRU P9. c) Standard deviation of interannual fluctuation δP = P − P

9, in964

AMIP (open dots), pre-industrial control (open squares) and historical (filled965

squares) experiments. The standard deviation has been normalized by the966

mean P 9. 49967

3 Precipitation (shaded, mm day−1) and 2-meter temperature (contours, K)968

bias of historical simulations against the corresponding AMIP simulations. 50969

44

4 Scatterplot of the ITCZ position against the meridional temperature gradient970

between the Gulf of Guinea and Sahara desert. The ITCZ latitude corresponds971

to the maximum of position of maximum precipitation averaged over 10◦W-972

10◦E. The temperature gradient is computed as the difference between the973

domains [20◦N-30◦N, 10◦W-10◦E] and [5◦S-5◦N, 10◦W-10◦E]. 51974

5 Precipitation (in mm day−1) averaged over 10◦W-10◦E, for the JAS period of975

the years 1979-2008 for CMIP5 simulations, and 1997-2008 for GPCP, 1998-976

2008 for TRMM, 1979–2002 for CRU and 1979–2008 for CMAP datasets.977

Each dataset is reproduced with its own horizontal resolution. 52978

6 2-meter temperature (in K) averaged over 10◦W-10◦E, for the JAS period979

of the years 1979-2008 for CMIP5 simulations, reanalyses and CRU. Each980

dataset is reproduced with its own horizontal resolution. 53981

7 Latitude-height diagrams of the cloud fraction averaged between 10◦W-10◦E,982

for the period JAS of the years 2006-2009 for the CloudSat-CALIPSO data983

set and 1989-2008 for the models. 54984

8 a) Cloud radiative effect (in W m−2) at the top of the atmosphere in the SW985

(negative values) and LW (positive values) bands, averaged over 10◦W-10◦E,986

for the JAS period of the years 1979-2008. b) As in a), but for the net CRE987

at the top of the atmosphere. 55988

9 Annual cycle of precipitation (in mm day−1 averaged over 10◦W-10◦E. A 10-989

day running mean was performed on each dataset. 56990

10 Annual cycle of 2-meter temperature (in K) averaged over [10◦N-20◦N,10◦W-991

10◦E]. 57992

11 Variance of OLR in the 1-90-day band, in W2 m−4 for the JAS periods from993

1979 to 2008. Daily OLR of CMIP5 models was regridded on the NOAA OLR994

grid (2.5◦x2.5◦), before computing filtered anomalies and their variance. 58995

45

12 a) Variance of OLR in the 1-90-day band averaged over the domain [5◦N-20◦N,996

10◦W-10◦E], and for the JAS periods from 1979 to 2008. Values are normal-997

ized by the NOAA OLR variance. b) Distribution of the OLR intraseasonal998

variance (in %) across the 1-3-day, 3-10-day and 10-90-day bands. 59999

13 a) Autocorrelation of a) JAS and b) DJF 1-90-day filtered precipitation at a 1-1000

day lag for the GPCP daily dataset. Only grid-point where mean precipitation1001

is greater than 1 mm day−1 are considered. c) Autocorrelation of 1-90-day1002

filtered precipitation at a 1-day lag for GPCP and CMIP5 models, averaged1003

over the domain [5◦N-15◦N,10◦W-10◦E]. Autocorrelation is computed for each1004

JAS season of the period 1997-2008 for GPCP and 1979–2008 for CMIP51005

model, and then averaged over all years. CMIP5 models and the GPCP1006

dataset were regridded on the NOAA OLR grid before any computation. 601007

14 Diurnal cycle of frequency-intensity distribution of precipitation and intensity1008

distribution (including null value) for August of the years 1999-2011 for the1009

rain gauge data and 1989-2008 for the models at the Ouémé site (2◦E-9.5◦N)1010

based on the 30-min samples. The black line superimposed on the distribution1011

is the mean diurnal rain intensity. For data, gray contours and squares su-1012

perimposed on the distributions are the same but using the 3-hourly samples1013

for comparison with the CNRM-CM5 model. For comparison the same figure1014

obtained for the Niamey site (2.2◦E-13.5◦N) is also shown. 611015

15 Diurnal cycle of cloud frequency of occurrence derived from AMF data and dis-1016

tribution of cloud fraction normalized by level with the total cloud frequency1017

of occurrence superimposed in dashed line for August 2006 and for August of1018

the years 1989-2009 for the models at the Niamey site (2.2◦E-13.5◦N). 621019

46

16 a) Downward longwave radiative flux at the surface (in W m−2), averaged1020

over 10◦W-10◦E, for the JAS period of the years 1979-2008. b) As in a), but1021

for the downward shortwave radiative flux at the surface. Mean fluxes for the1022

sites along the transect (Table 3) together with their yearly minimum and1023

maximum values are indicated. 631024

17 a) Scatterplot of the downward longwave radiative flux at the surface (in W1025

m−2) against the 2-meter temperature (in K). Both variables are averaged over1026

[20◦N-30◦N, 10◦W-10◦E], for the JAS period of the years 1979-2008. b) As in1027

a), but for the downward shortwave radiative flux at the surface against the1028

2-meter temperature. c) as in a) but for the downward longwave radiative flux1029

at the surface against the downward shortwave radiative flux at the surface (in1030

W m−2). The black square with error bars indicates the mean obtained from1031

observational datasets, i.e. SRB3pr, SRB3qc and CERES-EBAF for radiative1032

fluxes, and ERAI, MERRA and CFSR for the 2-meter temperature. 641033

47

a) dPr vs dT [10N-18N,13W-10E] - JAS

0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0

0

80

40

140

-20

120

20

-40

100

60

CMIP3 - SRESA2

CMIP5 - RCP85

CMIP5 - RCP45

dP

re

cip

[10

N-1

8N

,13

W-1

0E

](%

)

dT [10N-18N,13W-10E] (K)

b) dT [10N-18N,13W-10E] vs dT [global] - JAS

0 1 2 3 4 5 60

1

2

3

4

5

6

7

CMIP3 - SRESA2

CMIP5 - RCP85

CMIP5 - RCP45

y=1.06x+0.46 (R=0.69)

dT

[10

N-1

8N

,13

W-1

0E

](K

)

dT [global] (K)

c) dT [10N-18N,13W-10E] vs dT [global] - MAM

0 1 2 3 4 5 60

1

2

3

4

5

6

7

CMIP3 - SRESA2

CMIP5 - RCP85

CMIP5 - RCP45

y=1.28x+0.37 (R=0.90)

dT

[10

N-1

8N

,13

W-1

0E

](K

)

dT [global] (K)

Fig. 1. a) Climate projections of precipitation (in %) plotted against those of 2-metertemperature (in K) averaged over a Sahel domain [10◦N-18◦N, 13◦W-10◦E]. The differenceare computed between period 2071–2100 and the period 1971–2000, for the JAS season, forCMIP3 SRESA2 scenario and for CMIP5 RCP4.5 and RCP8.5 scenarios. b) Climate projec-tions of global 2-meter temperature plotted against those of 2-meter temperature averagedover the Sahel domain [10◦N-18◦N, 13◦W-10◦E]. Red points indicate projections for whichprecipitation change is greater than 15%. c) Same as b), but for the MAM season.

48

a)

1900 1920 1940 1960 1980 2000Year

3

4

5

6

7

Sah

el J

AS

rai

nfa

ll (

mm

/day

)

CRUCMAPCRU(9yr run. mean)

CMAP(9yr run. mean)

b) c)

-10 -5 0 5 10 15 20 25Rainfall decadal trend (amip) and std. deviation (%)

ACCESS1-3

BCC-CSM1-1

CCSM4

CESM1-CAM5

CNRM-CM5

CSIRO-Mk3-6-0

FGOALS-g2

FGOALS-s2

GISS-E2-R

IPSL-CM5A-LR

IPSL-CM5B-LR

MIROC5

MPI-ESM-LR

MPI-ESM-MR

MRI-CGCM3

NorESM1-M

OBSamip

historicalpiControl

0 5 10 15 20 25 30Rainfall interannual std deviation (%)

ACCESS1-3

BCC-CSM1-1

CCSM4

CESM1-CAM5

CNRM-CM5

CSIRO-Mk3-6-0

FGOALS-g2

FGOALS-s2

GISS-E2-R

IPSL-CM5A-LR

IPSL-CM5B-LR

MIROC5

MPI-ESM-LR

MPI-ESM-MR

MRI-CGCM3

NorESM1-M

OBSamip

historicalpiControl

Fig. 2. a) Time evolution of raw (thin line with dots) precipitation in mm day1 averagedover [10◦N-18◦N, 13◦W-10◦E] and of its decadal component P 9 (thick solid line), computedas the 9-year running mean of the raw index. CRU data is indicated by the thick black lineand CMAP by the thin black line. b) Mean difference of P 9 between the 199-2004 and 1983–1988 periods for AMIP simulations and CRU precipitations (dots), and standard deviationof P 9 in pre-industrial control (open squares) and historical (filled squares) experiments.The standard deviation has been normalized by the mean P 9. For observation, the filled redsquare corresponds to the standard deviation of CRU P

9. c) Standard deviation of interan-nual fluctuation δP = P − P

9, in AMIP (open dots), pre-industrial control (open squares)and historical (filled squares) experiments. The standard deviation has been normalized bythe mean P 9.

49

Fig. 3. Precipitation (shaded, mm day−1) and 2-meter temperature (contours, K) bias ofhistorical simulations against the corresponding AMIP simulations.

50

ITCZ latitude vs SHL-GG T2m gradient [10W-10E]

0.0 2.0 4.0 6.0 8.0 10.0 12.0 14.00

2

4

6

8

10

12

14

16

18CMIP5 - AMIP

CMIP3 - AMIP

CMIP5 - Historical

CMIP3 - 20c3m

R=0.51

T2m

[20N-30N]-[5S-5N](K

)

ITCZ Latitude (degrees N)

Fig. 4. Scatterplot of the ITCZ position against the meridional temperature gradient be-tween the Gulf of Guinea and Sahara desert. The ITCZ latitude corresponds to the maximumof position of maximum precipitation averaged over 10◦W-10◦E. The temperature gradientis computed as the difference between the domains [20◦N-30◦N, 10◦W-10◦E] and [5◦S-5◦N,10◦W-10◦E].

51

JAS Precipitation Climatology [10W-10E]

0 5N 10N 15N 20N 25N5S0

2

4

6

8

10

12

14

16bcc-csm1-1

GPCP

TRMM

CRU

CMAP

CCSM4

CNRM-CM5

CSIRO-Mk3-6-0

CanAM4

FGOALS-s2

GFDL-HIRAM-C180

GFDL-HIRAM-C360

HadGEM2-A

inmcm4

IPSL-CM5A-LR

IPSL-CM5B-LR

MPI-ESM-LR

MPI-ESM-MR

MIROC5

MRI-CGCM3

NorESM1-M

Pre

cip

ita

tio

n(m

m/d

ay

)

Latitude

Fig. 5. Precipitation (in mm day−1) averaged over 10◦W-10◦E, for the JAS period of theyears 1979-2008 for CMIP5 simulations, and 1997-2008 for GPCP, 1998-2008 for TRMM,1979–2002 for CRU and 1979–2008 for CMAP datasets. Each dataset is reproduced with itsown horizontal resolution.

52

JAS T2m Climatology [10W-10E]

0 5N 10N 15N 20N 25N 30N 35N 40N

296

298

300

302

304

306

308

310

312

314bcc-csm1-1

CFSR

MERRA

ERAI

CRU

CCSM4

CNRM-CM5

CSIRO-Mk3-6-0

CanAM4

FGOALS-s2

GFDL-HIRAM-C180

GFDL-HIRAM-C360

HadGEM2-A

inmcm4

IPSL-CM5A-LR

IPSL-CM5B-LR

MPI-ESM-LR

MPI-ESM-MR

MIROC5

MRI-CGCM3

NorESM1-M

T2

m(K

)

Latitude

Fig. 6. 2-meter temperature (in K) averaged over 10◦W-10◦E, for the JAS period of theyears 1979-2008 for CMIP5 simulations, reanalyses and CRU. Each dataset is reproducedwith its own horizontal resolution.

53

Fig. 7. Latitude-height diagrams of the cloud fraction averaged between 10◦W-10◦E, forthe period JAS of the years 2006-2009 for the CloudSat-CALIPSO data set and 1989-2008for the models.

54

a) JAS Climatology of LW/SW CRE at TOA

0 5N 10N 15N 20N 25N 30N 35N 40N10S 5S

0

20

40

60

80

100

120

-120

-100

-80

-60

-40

-20

bcc-csm1-1

CERES-EBAF

SRB3

bcc-csm1-1

CCSM4

CNRM-CM5

CSIRO-Mk3-6-0

CanAM4

FGOALS-s2

GFDL-HIRAM-C180

GFDL-HIRAM-C360

HadGEM2-A

inmcm4

IPSL-CM5A-LR

IPSL-CM5B-LR

MPI-ESM-LR

MPI-ESM-MR

MIROC5

MRI-CGCM3

NorESM1-Mbcc-csm1-1

CERES-EBAF

SRB3

bcc-csm1-1

CCSM4

CNRM-CM5

CSIRO-Mk3-6-0

CanAM4

FGOALS-s2

GFDL-HIRAM-C180

GFDL-HIRAM-C360

HadGEM2-A

inmcm4

IPSL-CM5A-LR

IPSL-CM5B-LR

MPI-ESM-LR

MPI-ESM-MR

MIROC5

MRI-CGCM3

NorESM1-M

LW CRE

SW CRE

CR

E(W

/m2)

Latitudeb) JAS Climatology of Net CRE at TOA

0 5N 10N 15N 20N 25N 30N 35N 40N10S 5S

0

20

40

60

-100

-80

-60

-40

-20

bcc-csm1-1

CERES-EBAF

SRB3

bcc-csm1-1

CCSM4

CNRM-CM5

CSIRO-Mk3-6-0

CanAM4

FGOALS-s2

GFDL-HIRAM-C180

GFDL-HIRAM-C360

HadGEM2-A

inmcm4

IPSL-CM5A-LR

IPSL-CM5B-LR

MPI-ESM-LR

MPI-ESM-MR

MIROC5

MRI-CGCM3

NorESM1-M

CR

E(W

/m2)

Latitude

Fig. 8. a) Cloud radiative effect (in W m−2) at the top of the atmosphere in the SW(negative values) and LW (positive values) bands, averaged over 10◦W-10◦E, for the JASperiod of the years 1979-2008. b) As in a), but for the net CRE at the top of the atmosphere.

55

0

1

3

5

7

9

11

13

a) bcc-csm1-1

0

10N

20N

10S

b) CanAM4 c) CNRM-CM5 d) CSIRO-Mk3-6-0

e) FGOALS-g2

0

10N

20N

10S

f) FGOALS-s2 g) GFDL-HIRAM-C180 h) GFDL-HIRAM-C360

i) HadGEM2-A

0

10N

20N

10S

j) inmcm4 k) IPSL-CM5A-LR l) IPSL-CM5B-LR

m) MIROC5

0

10N

20N

10S

n) MPI-ESM-LR o) MPI-ESM-MR p) MRI-CGCM3

q) NorESM1-M

J DNOSAJJMAMF

0

10N

20N

10S

r) TRMM

J DNOSAJJMAMF

s) GPCP

J DNOSAJJMAMF

Fig. 9. Annual cycle of precipitation (in mm day−1 averaged over 10◦W-10◦E. A 10-dayrunning mean was performed on each dataset.

56

Fig. 10. Annual cycle of 2-meter temperature (in K) averaged over [10◦N-20◦N,10◦W-10◦E].

57

0

200

400

600

800

1000

1200

1400

1600

1800

a) bcc-csm1-1

0

40N

10N

20N

10S

30N

b) CanAM4 c) CNRM-CM5 d) CSIRO-Mk3-6-0

e) FGOALS-g2

0

40N

10N

20N

10S

30N

f) FGOALS-s2 g) GFDL-HIRAM-C180 h) GFDL-HIRAM-C360

i) HadGEM2-A

0

40N

10N

20N

10S

30N

j) inmcm4 k) IPSL-CM5A-LR l) IPSL-CM5B-LR

m) MIROC5

0

40N

10N

20N

10S

30N

n) MPI-ESM-LR o) MPI-ESM-MR p) MRI-CGCM3

q) NorESM1-M

0 20E 40E40W 20W

0

40N

10N

20N

10S

30N

r) NOAA

0 20E 40E40W 20W

Fig. 11. Variance of OLR in the 1-90-day band, in W2 m−4 for the JAS periods from 1979to 2008. Daily OLR of CMIP5 models was regridded on the NOAA OLR grid (2.5◦x2.5◦),before computing filtered anomalies and their variance.

58

0  

0,2  

0,4  

0,6  

0,8  

1  

1,2  

1,4  

1,6  

GFDL-HIRAM-C

180

NorESM1-M

GFDL-HIRAM-C

360

MRI-CGCM3

CNRM-CM5

bcc-csm1-1

MPI-ESM-L

R

MPI-ESM-M

R

CSIRO-Mk3-6-0

CanAM4

FGOALS-g2

MIROC5

FGOALS-s2

HadGEM2-A

inmcm4

IPSL-CM5B-L

R

IPSL-CM5A-L

R

NOAA

Varia

nce  ra)o

 to  NOAA

 

a)    Ra)o  of  model  OLR  intraseasonal  variance  to  that  of  NOAA  OLR  [5N-­‐20N,  10W-­‐10E]  

0  

10  

20  

30  

40  

50  

60  

70  

GFDL-HIRAM-C

180

NorESM1-M

GFDL-HIRAM-C

360

MRI-CGCM3

CNRM-CM5

bcc-csm1-1

MPI-ESM-L

R

MPI-ESM-M

R

CSIRO-Mk3-6-0

CanAM4

FGOALS-g2

MIROC5

FGOALS-s2

HadGEM2-A

inmcm4

IPSL-CM5B-L

R

IPSL-CM5A-L

R

NOAA

Scale  contrib

u.on

 (%)  

b)  Distribu.on  of  OLR  intraseasonal  variance  [5N-­‐20N,  10W-­‐10E]  

10d  <  T  <  90d   3d  <  T  <  10d   T  <  3d  

Fig. 12. a) Variance of OLR in the 1-90-day band averaged over the domain [5◦N-20◦N,10◦W-10◦E], and for the JAS periods from 1979 to 2008. Values are normalized by theNOAA OLR variance. b) Distribution of the OLR intraseasonal variance (in %) across the1-3-day, 3-10-day and 10-90-day bands.

59

a) Autocorrelation of GPCP precip anomalies at lag 1 day - JAS

030W60W90W180W 120W 180W150E120E90E60E150W 30E

0

40N

20S

20N

40S

-1 -0.8 -0.6 -0.4 -0.2 -0.1 0 0.1 0.2 0.4 0.6 0.8 1

b) Autocorrelation of GPCP precip anomalies at lag 1 day - DJF

030W60W90W180W 120W 180W150E120E90E60E150W 30E

0

40N

20S

20N

40S

-­‐0,05  0  

0,05  0,1  0,15  0,2  0,25  0,3  0,35  0,4  0,45  0,5  0,55  0,6  

GFDL-HIRAM-C

180

NorESM1-M

GFDL-HIRAM-C

360

MRI-CGCM3

CNRM-CM5

bcc-csm1-1

MPI-ESM-L

R

MPI-ESM-M

R

CSIRO-Mk3-6-0

CanAM4

FGOALS-g2

MIROC5

FGOALS-s2

HadGEM2-A

inmcm4

IPSL-CM5B-L

R

IPSL-CM5A-L

R

GPCP

Autocorrela*

on  at  1

-­‐day  lag  

c)    Autocorrela*on  of  precipita*on  intraseasonal  anomalies  at  1-­‐day  lag  [5N-­‐15,  10W-­‐10E]  

Fig. 13. a) Autocorrelation of a) JAS and b) DJF 1-90-day filtered precipitation at a 1-day lag for the GPCP daily dataset. Only grid-point where mean precipitation is greaterthan 1 mm day−1 are considered. c) Autocorrelation of 1-90-day filtered precipitation at a1-day lag for GPCP and CMIP5 models, averaged over the domain [5◦N-15◦N,10◦W-10◦E].Autocorrelation is computed for each JAS season of the period 1997-2008 for GPCP and1979–2008 for CMIP5 model, and then averaged over all years. CMIP5 models and theGPCP dataset were regridded on the NOAA OLR grid before any computation.

60

Fig. 14. Diurnal cycle of frequency-intensity distribution of precipitation and intensitydistribution (including null value) for August of the years 1999-2011 for the rain gauge dataand 1989-2008 for the models at the Ouémé site (2◦E-9.5◦N) based on the 30-min samples.The black line superimposed on the distribution is the mean diurnal rain intensity. Fordata, gray contours and squares superimposed on the distributions are the same but usingthe 3-hourly samples for comparison with the CNRM-CM5 model. For comparison the samefigure obtained for the Niamey site (2.2◦E-13.5◦N) is also shown.

61

Fig. 15. Diurnal cycle of cloud frequency of occurrence derived from AMF data and dis-tribution of cloud fraction normalized by level with the total cloud frequency of occurrencesuperimposed in dashed line for August 2006 and for August of the years 1989-2009 for themodels at the Niamey site (2.2◦E-13.5◦N).

62

a) JAS Climatology of LWdn at surface

0 5N 10N 15N 20N 25N 30N 35N 40N10S 5S320

340

360

380

400

420

440

460

480bcc-csm1-1

SRB3pr

SRB3qc

CERES-EBAF

bcc-csm1-1

CCSM4

CNRM-CM5

CSIRO-Mk3-6-0

CanAM4

FGOALS-s2

GFDL-HIRAM-C180

GFDL-HIRAM-C360

HadGEM2-A

inmcm4

IPSL-CM5A-LR

IPSL-CM5B-LR

MPI-ESM-LR

MPI-ESM-MR

MIROC5

MRI-CGCM3

NorESM1-MS

urfa

ce

LW

dn

(W/m

2)

Latitudeb) JAS Climatology of SWdn at surface

0 5N 10N 15N 20N 25N 30N 35N 40N10S 5S

150

200

250

300

350

400bcc-csm1-1

SRB3pr

SRB3qc

CERES-EBAF

bcc-csm1-1

CCSM4

CNRM-CM5

CSIRO-Mk3-6-0

CanAM4

FGOALS-s2

GFDL-HIRAM-C180

GFDL-HIRAM-C360

HadGEM2-A

inmcm4

IPSL-CM5A-LR

IPSL-CM5B-LR

MPI-ESM-LR

MPI-ESM-MR

MIROC5

MRI-CGCM3

NorESM1-M

Latitude

Fig. 16. a) Downward longwave radiative flux at the surface (in W m−2), averaged over10◦W-10◦E, for the JAS period of the years 1979-2008. b) As in a), but for the downwardshortwave radiative flux at the surface. Mean fluxes for the sites along the transect (Table3) together with their yearly minimum and maximum values are indicated.

63

a) T2m vs LWdn at Surface [20N-30N,10W-10E]

420.0360.0 400.0340.0 380.0302

303

304

305

306

307

308

309

310R=0.82

T2

m(K

)

LWdn at surface (W/m2)

b) T2m vs SWdn at Surface [20N-30N,10W-10E]

320260 300240 340280220302

303

304

305

306

307

308

309

310R=0.49

T2

m(K

)

SWdn at surface (W/m2)

c) SWdn vs LWdn at Surface [20N-30N,10W-10E]

420360 400340 380

320

260

300

240

340

280

220

bcc-csm1-1

CCSM4

CNRM-CM5

CSIRO-Mk3-6-0

CanAM4

FGOALS-s2

GFDL-HIRAM-C180

GFDL-HIRAM-C360

HadGEM2-A

inmcm4

IPSL-CM5A-LR

IPSL-CM5B-LR

MPI-ESM-LR

MPI-ESM-MR

MIROC5

MRI-CGCM3

NorESM1-M

R=0.73

SW

dn

at

su

rfa

ce

(W/m

2)

LWdn at surface (W/m2)

Fig. 17. a) Scatterplot of the downward longwave radiative flux at the surface (in Wm−2) against the 2-meter temperature (in K). Both variables are averaged over [20◦N-30◦N,10◦W-10◦E], for the JAS period of the years 1979-2008. b) As in a), but for the downwardshortwave radiative flux at the surface against the 2-meter temperature. c) as in a) butfor the downward longwave radiative flux at the surface against the downward shortwaveradiative flux at the surface (in W m−2). The black square with error bars indicates themean obtained from observational datasets, i.e. SRB3pr, SRB3qc and CERES-EBAF forradiative fluxes, and ERAI, MERRA and CFSR for the 2-meter temperature.

64