Climate Dynamics

, Volume 48, Issue 7–8, pp 2751–2770 | Cite as

On the range of future Sahel precipitation projections and the selection of a sub-sample of CMIP5 models for impact studies

  • Paul-Arthur Monerie
  • Emilia Sanchez-Gomez
  • Julien Boé


The future evolution of the West African Monsoon is studied by analyzing 32 CMIP5 models under the rcp8.5 emission scenario. A hierarchical clustering method based on the simulated pattern of precipitation changes is used to classify the models. Four groups, which do not agree on the simple sign of future Sahel precipitation change, are obtained. We find that the inter-group differences are mainly associated with the large spread in (1) temperature increase over the Sahara and North Atlantic and in (2) the strengthening of low and mid-level winds. A wetter Sahel is associated with a strong increase in temperature over the Sahara (>6 °C), a northward shift of the monsoon system and a weakening of the African Easterly jet. A dryer Sahel is associated with subsidence anomalies, a strengthening of the 600 hPa wind speed, and a weaker warming over the Northern Hemisphere. Moreover, the western (central) Sahel is projected to become dryer (wetter) during the first months (last months) of the rainy season in a majority of models. We propose several methods to select a sub-sample of models that captures both the ensemble mean pattern and/or the spread of precipitation changes from the full ensemble. This methodology is useful in all the situations for which it is not possible to deal with a large ensemble of models, and in particular most impact studies. We show that no relationship exists between the climatological mean biases in precipitation and temperature and the future changes in the monsoon intensity. This indicates that the mean bias is therefore not a reliable metric for the model selection. For this reason, we propose several methodologies, based on the projected precipitation changes: The “diversity” method, which consists in the selection of one model from each group is the most appropriate to capture the spread in precipitation change. The “pattern selection” method, which consists in the selection of models in a single group allows to select models for the study of a specific pattern of precipitation change, for example the one that is the most representative of the full ensemble.

1 Introduction

Observational studies show that the 1970s and 1980s were abnormally dry in the Sahel (L’Hote et al. 2002, 2003; Dai et al. 2004; Nicholson et al. 2012a, b). Since the 1990s, Sahel rainfall has experienced a limited precipitation recovery (Ozer et al. 2003; Nicholson 2005; Lebel and Ali 2009), especially from August to October (Sanogo et al. 2015). A recent study based on numerical modeling suggests that the recovery is due to changes in atmospheric greenhouse gases (GHGs) and aerosols concentration (Dong and Sutton 2015). It could thus be expected that the recent Sahel precipitation intensification might continue as the GHGs concentration increases. Based on studies using the CMIP3 and CMIP5 databases (Coupled Model Intercomparison Phase 3 and 5), the 4th and 5th IPCC (Intergovernmental Panel on Climate Change) assessment reports however concluded that a high level of uncertainty in future changes in Sahel climate exists (Solomon et al. 2007; Barros et al. 2015).

Sahel precipitation is inextricably linked to the West Africa Monsoon (WAM). Studies based on CMIP3/5 projections show that the WAM response to global warming is highly model-dependent. As a consequence, model based projections of WAM and Sahel climate are highly uncertain. This is particularly problematic for the assessment of socio-economic impacts, since in order to capture correctly the uncertainties (model formulation, internal variability, emission scenario), a large ensemble of models is necessary, as well as different realizations and scenario emissions should be considered. The CMIP5 multi-model ensemble consists in more than 40 climate models, with generally several realizations (or members), and for 4 different Representative Concentration Pathways (RCPs). It is virtually impossible for most groups working on dynamical downscaling and/or impacts, given limitations in computing resources and/or data storage capacity, to deal with such a large ensemble. Sub-sampling is necessary and given the major inter-model differences regarding the future evolution of Sahel precipitation, it is particularly complex in this context.

The need for an adequate sub-ensemble of models is becoming more and more frequent, since in the recent years an increasing number of multi-disciplinary impact studies have emerged: they are based on crop models (Sultan et al. 2014); on hydrological models (Li et al. 2015); on land-cover and land use models (Xue et al. 2012); on malaria (Caminade et al. 2014) and meningitis modeling (Abdussalam et al. 2014); among others. The African countries will be indeed affected by global warming in many areas and will face important adaptation challenges. In this study, we propose different approaches to define an optimal sub-sample of models (composed by a small number of models, which is able to provide a mean pattern and inter-model spread of precipitation change close to them of the CMIP5 ensemble), based on the models response in Sahel precipitation under global warming.

The first step of the study is focused on the analysis of the diversity of CMIP5 future projections for Sahel precipitation. Precipitation is projected to increase (decrease) over the central (western) Sahel in both CMIP3 and CMIP5 multi-model ensembles (Fontaine et al. 2011b; Monerie et al. 2012, 2013; James and Washington 2013; Biasutti 2013; James et al. 2015). In a recent study, Park et al. (2015) suggest that the discrepancies in future Sahel projections can be explained by differences between Northern and Southern Hemisphere surface warming. Here, we revisit this question with an original statistical approach, based on hierarchical clustering applied to the projected patterns of Sahel precipitation change.

Sahel precipitation change uncertainty is caused by multiple competing physical mechanisms that impact the interannual-to-multidecadal variability of Sahel precipitation. Additionally, the Tropical Atlantic is a region of large model uncertainties: state-of-the-art climate models exhibit large systematic errors (Richter and Xie 2008; Richter et al. 2012) and large uncertainties exist in the relative roles of internal and external factors in shaping climate change. The review of Druyan (2011) highlights that the studies based on CMIP3 projections often show opposite results, ranging from a significant increase to a significant decrease in Sahel precipitation at the end of the twenty first century. Different physical mechanisms have been proposed to explain the opposite responses found in coupled models. On the one hand, a wetter Sahel is associated with increased land-sea temperature contrast (Haarsma et al. 2005) between the Gulf of Guinea and the Sahara, and with a strengthening of the Tropical North Atlantic/Tropical South Atlantic temperature gradient (Hoerling et al. 2006; Park et al. 2015). On the other hand, the Tropical Ocean warming may induce an increase in moist static energy at upper levels, affecting then the vertical stability of the atmosphere, leading to a drying over the Sahel, in a way analogous to the impacts of El Niño Southern Oscillation on the global tropical atmosphere (Giannini et al. 2010). Our objective is to better understand the physical mechanisms responsible for the major inter-model spread in Sahel precipitation change. We also investigate whether the future model responses are linked to present-day climatological biases in Sahel precipitation. This question is of interest, since some models sub-sampling approaches are based on the models ability to simulate the mean historical climate (as in Buontempo et al. 2015), making the implicit hypothesis that models with smaller biases provide more reliable future projections.

In a second step, we test several methods for the selection of a subset of CMIP5 models in the context of Sahel precipitation changes. The main objective is to find a small subset of models that both reproduces the ensemble mean and inter-model spread of the full CMIP5 ensemble.

The main scientific questions addressed in this study can be summarized as follows:
  1. 1.

    Which are the main responses in future Sahel precipitation and the associated physical mechanism in CMIP5 models at the monthly time-scale?

  2. 2.

    Is the large inter-model spread in future precipitation changes linked to present-day climate characteristics?

  3. 3.

    Is it possible to define a subset of models, representative of the full ensemble in terms of ensemble mean and spread?


The paper is organized as follows: In Sect. 2 we describe the methodology and the data used. The mean changes in Sahel rainfall are analyzed in Sect. 3. Section 4 focuses on models biases and proposes different methods for model selection. Finally we conclude in Sect. 5 with a summary and a discussion of our results.

2 Data and methodology

2.1 Data

Observed precipitation for the 1979–2010 period from the GPCP v2.2 data set is used with a 2.5° × 2.5° horizontal grid resolution. GPCP is a merged analysis that incorporates precipitation estimates from microwave data, infrared data, and surface rain gauge observations (Adler et al. 2003). Surface temperature, wind and sea level pressure for the period 1979–1999 come from the ERA-Interim (ERAI) reanalysis (Dee et al. 2011). Compared to ERA-40 (Uppala et al. 2005), ERAI provides an improved representation of the hydrological cycle over the tropics (Uppala et al. 2008), a more realistic stratospheric circulation and better temporal consistency of the reanalysis fields (Dee et al. 2011).

Two sets of numerical experiments from the CMIP5 archive (Taylor et al. 2012) are analyzed: the historical ensemble (hereinafter HIST, forced by both historical natural and anthropogenic forcings); and the RCP8.5 ensemble (noted RCP85, based on the most extreme scenario whose radiative forcing reaches 8.5 W/m2 in 2100; see Meinshausen et al. 2011). We selected the models for which the necessary variables to this study are available for both HIST and RCP85, which leads to an ensemble of 32 models (see the Fig. 1 for the model names).
Fig. 1

The model “family tree” from CMIP5 models for the RCP85-HIST change in Sahel precipitation, shown as a dendrogram (a hierarchical clustering of the pairwise distance matrix for precipitation, see text). Models on the same branch simulate similar patterns in precipitation changes. Models from the same group share the same color

In this work, mean model biases are estimated from the JAS (July–August–September) differences between HIST and GPCP or ERAI on the 1979–1999 period. The models response to global warming is defined as the difference between RCP85 in the 2060–2099 period and HIST on the 1960–1999 period. Models outputs, GPCP precipitation and ERAI variables have been interpolated on the same 2.5° latitude by 2.5° longitude grid to facilitate the analysis. One member is considered for each of the 32 models for both the HIST and RCP85 experiments. Throughout this work, the multi-model response is considered as robust if at least 80 % of the models agree on the sign of the change.

2.2 Model classification

Future projected changes in Sahel precipitation from the 32 CMIP5 models are classified with a hierarchical clustering algorithm (Jain et al. 1999). The spatial pattern to be classified is the JAS RCP85—HIST differences in precipitation over a Sahelian box defined as 20°West–20°East and 10°North–20°North (including both the Western and Central box in Fig. 2a). The spatial correlation matrix is then computed (32 × 32 size). The hierarchical clustering is performed following the ward minimum variance method (Ward 1963), based on the spatial correlation amongst the models. The position at which two models are connected on the tree characterizes their disagreement. Four groups of models are empirically defined based on the classification shown in Fig. 1, noted GR1, GR2, GR3 and GR4 hereinafter. The semi-empirical choice of the number of clusters rests on a compromise, aiming to avoid clusters of very small size (whose properties could be strongly impacted by an outlier), and to have a sufficient number of groups to capture correctly the discrepancies in Sahel precipitation changes.
Fig. 2

Projected changes in JAS ad precipitation (mm day−1), eh evaporation (mm day−1) and il P–E budget (mm day−1) for the four groups of models (one group per column) identified by the clustering. The present climatology (HIST) is displayed with red contours and the RCP85-HIST differences by color shading. Hatching represents the grid-points where at least 80 % of the models agree with the sign of the intra-group ensemble mean within each groups (5 for GR1; 7 for GR2; 13 for GR3; 5 for GR4)

Precipitation changes are expected to be consistent within each cluster, which is not the case for the full ensemble. The goal of this classification is to facilitate the understanding of the diversity of projected precipitation changes in the Sahel by CMIP5 models.

Two outliers are identified by the classification: two GFDL models (gfdl_esm2g and gfdl_esm2m), whose responses are largely different from the other models, and finally they are included in any group. As in Pennell and Reichler (2011), Masson and Knutti (2011) and Knutti et al. (2013), models from the same institution generally belong to the same cluster (i.e. HADGEM, GFDL, IPSL, CMCC, MIROC, GISS, ACCESS and MPI). These similarities are due to the fact that models from the same center sometimes only differ in resolution (ipsl_cm5a_mr and ipsl_cm5a_lr for example), and generally share large portions of numerical code or even identical component (e.g. ocean, atmosphere or land models) (Masson and Knutti 2011).

In the following section, changes in precipitation, temperature, pressure and winds over West Africa for the four groups of models defined above are studied.

3 Characterization of projected changes in the Sahel

3.1 Water budget

Future changes in JAS Sahel precipitation (P), evaporation (E) and P–E budget for the four groups of models identified by hierarchical clustering are depicted in Fig. 2. The intra-cluster mean is computed and the result is considered as robust when at least 80 % of models within the cluster agree on the sign of the change. GR1 exhibits a large increase in rainfall (>2 mm/day) over the entire northern Sahel (Fig. 2a). Over the adjacent ocean, evaporation slightly increases (Fig. 2e) yielding to a positive P-E budget, indicating an enhancement of moisture flux convergence (divergence), over the Sahel (Gulf of Guinea). This pattern has been widely documented in the past, and is associated with a strengthening of the land-ocean temperature contrast (Maynard et al. 2002; Haarsma et al. 2005; Skinner et al. 2012), and inhomogeneous ocean temperature changes within the Tropical Atlantic basin (i.e. the Tropical North Atlantic warms more than the Tropical South Atlantic, as reported in Hoerling et al. 2006). GR2 simulates a slight increase in precipitation over the central Sahel, and a decrease in precipitation over the Senegal (Fig. 2b). An important increase in precipitation along the maritime Inter-Tropical Convergence Zone (ITCZ) is also robust. GR2 also exhibits a strong increase in E over the adjacent ocean and west of central Africa (Fig. 2f). The resulting P–E budget is strongly negative over the Gulf of Guinea, and no robust changes are observed over the Sahel (Fig. 2j). Therefore, in GR2 the change in precipitation over the Sahel is not associated with a change in moisture flux convergence, suggesting that it might be linked to local changes in moisture recycling (the evaporation is moisture-limited rather than energy-limited in the semi-arid Sahel). GR3, which is the most populated cluster, shows a robust increase in P over the north-central Sahel and a decrease over its western part (Fig. 2c). The patterns for E and P changes are similar over the continent, but differ over the ocean, where E increases more strongly (Fig. 2c, g). The P–E budget increases (decreases) over the central (west) Sahel (Fig. 2k), indicating an increase (a decrease) in moisture flux convergence. These changes in the water budget have been already documented by Fontaine et al. (2011b), Monerie et al. (2012, 2013), James and Washington (2013), Biasutti (2013) and James et al. (2015): GR3 is the dominating response in the CMIP5 full-ensemble. The response for GR4 consists in a slight decrease in P over both the continent and the ocean (Fig. 2d). E decreases over the central Sahel but increases over the Gulf of Guinea area (Fig. 2 h). No robust changes are found for the P-E budget over the continent, whereas a robust decrease in P-E occurs over the Gulf of Guinea (Fig. 2l).

In summary, four main types of projected precipitation changes over the Sahel can be extracted from the CMIP5 full-ensemble: a strong increase in precipitation (>2 mm/day, GR1), a zonal dipole with an increase (a decrease) in precipitation over the western (central) part (GR2 and GR3, but with a small increase in precipitation over the central Sahel in GR2); and finally a decrease in P over the entire domain (GR4). These patterns are consistent with the large spread in the WAM changes simulated by CMIP3 models (Druyan, 2011), and illustrate the strong uncertainties in Sahel precipitation projections. In the next section we analyze the atmospheric dynamics associated with the P, E and P–E changes described above.

3.2 Atmospheric circulation and temperature

Figures 3a-d show the projected changes in 2 m temperature (T2m hereinafter, shading), sea level pressure (SLP, contours) and wind at 950 hPa (arrows) for the four groups of models. From Fig. 3, it is evident that the magnitude of T2m changes is different among the groups. However, the spatial patterns of warming present large similarities, with stronger increases in temperature over the continent than over the ocean, which act to increase the land-sea temperature gradients (Fig. 3a–d). The associated dynamics show a marked decrease in SLP over northern Africa, together with a strengthening of the south-westerlies (as shown in Haarsma et al. 2005) for the 4 groups of models. Over the Sahara, the SLP decreases, indicating a deeper heat low, which favors the strengthening of low-level winds. The main inter-group differences concern the intensity of the warming and of the wind strengthening.
Fig. 3

Projected changes RCP85-HIST in JAS (ad) 2 m temperature (°C) (shading), sea level pressure (hPa) (blue contours) and 950 hPa winds (m s−1) (green arrows) for the four groups of models. The winds anomalies are displayed if at least 80 % of the models agree on the sign of the intra-cluster mean. eh Latitude-height cross section (average from 10°W to 10°E) of zonal winds (m.s−1) for the RCP85-HIST projected changes (shading). The present climatology (HIST) for zonal winds is displayed with red contours. Hatching represents the grid-points where at least 80 % of the models agree with the sign of the intra-cluster mean (5 for GR1; 7 for GR2; 13 for GR3; 5 for GR4)

At global-scale, surface temperature increases more in the northern Hemisphere than in the Southern Hemisphere for the four groups, which could act to enhance precipitation over the Sahel (Park et al. 2015). This inter-hemispheric differential warming in JAS is considered here only over the Atlantic Ocean, and computed as the RCP85-HIST difference of SSTs averaged between 75°W and 10°E, minus the global tropical SST change (the global tropical domain is 180°W–180°E; 30°S–30°N). Results for the four groups of models are shown in Fig. 4. The major differences amongst the groups are located north of 40°N. GR1 and GR3 exhibit a stronger strengthening of the inter-hemispheric SST gradient over the Atlantic than GR2 and GR4. Interestingly, GR1 and GR3 are the groups that project an increase in Sahel precipitation (Fig. 2a, c). This result is consistent with Park et al. (2015), who have shown that the strengthening of this inter-hemispheric gradient is one of the main causes of the models discrepancies in Sahel precipitation changes.
Fig. 4

Changes in Atlantic SST gradient, computed as the zonally-averaged RCP85-HIST difference over the Atlantic Ocean (75°W–10°E) minus the SST tropical warming (180°W–180°E; 30°S–30°N) in JAS. The intra-cluster mean is computed for GR1 (brown line), GR2 (blue line), GR3 (green line) and GR4 (red line). Shadings represent the intra-cluster standard deviation

The low-level warming leads to stronger temperature gradients in GR1 and GR3 than in GR2 and GR4. This is consistent with a stronger increase in low-level winds and decrease in SLP, leading to larger increases in Sahel precipitation in GR1 and GR3 than in GR2 and GR4. The land-ocean temperature gradient increases in all the group of models and cannot explain the decrease in precipitation for GR4 and the very weak increase for GR2.

This difference in projected Sahel precipitation may be due to the upper-level wind dynamics, represented by the Tropical Easterly Jet (TEJ) and the African Easterly Jet (AEJ), which are known to influence Sahel precipitation (Nicholson 2008). A southward shift of the AEJ displaces the rain-belt southward, reducing the rainfall over the Sahel (Grist and Nicholson 2001) and the TEJ strength is significantly correlated with Sahel precipitation (Grist and Nicholson 2001). According to Cook (1999) and Patricola and Cook (2008), the AEJ can also reduce rainfall by transporting moisture away from the Sahel. The impact of the AEJ and TEJ on the precipitation change is however not straightforward. Precipitation impacts soil moisture and surface warming, leading to a change of the meridional gradients of surface moisture and temperature that displaces the AEJ (Thorncroft and Blackburn 1999). More intense precipitation also results in a stronger local deep meridional overturning circulation with upper-level northerlies that accelerate the TEJ. A feedback therefore exists between the AEJ (TEJ) and Sahel precipitation. We thus do not conclude on a direct impact of the jets on the precipitation change.

Following this, we investigate the low to upper levels dynamics considering the latitude-height cross-sections of the zonal wind component. The Sahel domain (from 10°W to 10°E) is selected to compute longitudinal averages. The intra-cluster mean for each group is shown in Fig. 3e–h. During the JAS rainy season, all the groups simulate a strengthening of the surface winds (from equator to 15°N and between 1000 to 850 hPa) (Fig. 3e–h). The strengthening of the westerlies is larger for GR1 and GR3, likely because of the greater warming over northern Africa (Fig. 3a, c). The low-level winds also accelerate in response to increased latent heating due to the increased rainfall and low-level moisture. At mid-level (600–400 hPa), the AEJ moves northward for GR1, as shown by the positive (negative) anomalies south (north) of the wind core at 600 hPa and 15°N (Fig. 3e). No robust changes in the AEJ speed are obtained for GR3 (Fig. 3g). This could be due to the choice of the cross-section, centered on the Greenwich meridian, whereas for GR3 there is a zonal dipole of precipitation change and thus likely, a non-homogeneous zonal response. The two others groups agree on an increase in the AEJ speed and on its southward shift (negative anomalies south of the AEJ core) (Fig. 3f, h).

At upper level, the TEJ, whose core is located generally at 200 hPa, exhibits a strengthening and a displacement southwards for GR1 and GR3 (Fig. 3e, g). For GR2 and GR4, the TEJ speed decreases at all latitudes range. This is consistent with the increase in precipitation, as shown previously in the literature (Nicholson 2008).

The projected responses in T2m, SLP and zonal wind vertical structure are generally consistent among the groups of models in terms of spatial pattern, but with differing intensities (Fig. 3). The main differences are associated with different changes in the vertically-integrated moisture flux convergence, represented by the P–E budget (Fig. 2i–l). The latter are due to the competing effects of the feeding of moisture through the low-levels winds (Fig. 3e–h) and the export of moisture through the mid-level winds (Fig. 3e–h). GR1 models simulate a northward shift of the monsoon flow, a weakening and a northward shift of the AEJ. On the contrary, GR2 and GR4 models simulate a strengthening of the AEJ, in consistency with a decrease in precipitation in GR4 and with no northward displacements of the monsoon system in GR2.

For GR3, there are no evident changes in mid-level winds. This group exhibits zonally contrasted changes in precipitation (increase/decrease in the central/western Sahel) and the box centered on the Greenwich meridian is not suitable for the analysis of this pattern. Therefore, in the following Sect. 3.3, we consider two areas to characterize the projected changes: the western and central Sahel (Fig. 2a, gray boxes).

3.3 Seasonal cycle

The increase in precipitation noted in Fig. 2a–d may be associated with a change in the seasonal cycle of the WAM. For example, Biasutti (2013) showed from an ensemble of CMIP5 models that a pattern of precipitation similar to the one of GR3 is associated with a decrease in precipitation over the western Sahel in June–July, and an increase in precipitation over the central Sahel in September–October. To investigate further this question, we analyze the latitude-seasonal cycle diagrams of the water budget, zonal winds at 600 hPa (AEJ) and 400 hPa vertical wind velocity over the western Sahel (20°W–0°E) and central Sahel (0°E–20°E) domains. Figures 5 and 6 show the projected changes in P (Figs. 5a–d, 6a–d) and P–E (Fig. 5e–h, Fig. 6e–h) over the western and central Sahel respectively. Zonal wind at 600 hPa (Figs. 5i–l, 6i–l) and 400 hPa vertical wind velocity (Figs. 5m–p, 6m–p) changes are also displayed in the same format.
Fig. 5

Time-Latitude diagram from January to December for RCP85-HIST projected changes averaged over the western Sahel (20°W–0°W) in (ad) precipitation (mm day−1), eh P–E (mm day−1), il 600 hPa zonal wind (m s−1) and (mp) 400 hPa omega (Pa s−1) for the four groups of models. The present climatology (HIST) is displayed with red contours and the HIST-RCP85 differences by the shading. Hatching represents the grid-points where at least 80 % of the models agree with the sign of the intra-cluster mean (5 for GR1; 7 for GR2; 13 for GR3; 5 for GR4)

Fig. 6

Same as the Fig. 5 but for the central Sahel (from 0° to 20°E)

First, we analyze the changes in the seasonal cycle over the western box (Fig. 5). GR1 simulates a strong increase in precipitation from July to October over the Sahel and in boreal winter south of the equator (Fig. 5a). This change is associated with an increase in moisture convergence (Fig. 5e), a monsoon system located northward in JAS, as shown by the northward shift of the AEJ (Fig. 5i), and with ascending (subsiding) anomalies north (south) of 10°N (Fig. 5m). GR2, GR3 and GR4 project a decrease in P over the Sahel (Fig. 5b–d), but the timing in P changes is different, varying from the March to August period for GR2 to the July to October period for GR4. These changes are associated with a decrease in moisture flux convergence (Fig. 5f–h) and with an increase in the AEJ strength (Fig. 5j–l). There is an evident subsiding anomaly for GR2, GR3, and GR4 (Fig. 5n–p). These findings obtained for GR2, GR3 and GR4 (i.e., a decrease in rainfall over western Sahel, subsidence anomalies and AEJ speed increases) are consistent with Monerie et al. (2012, 2013) and James et al. (2015).

On the central Sahel, an increase in precipitation is seen from July to December for GR1 (Fig. 6a), GR2 (Fig. 6b) and GR3 (Fig. 6c). GR4 exhibits a decrease in precipitation from April to October and an increase at the end of the year. A wetter Sahel during the late rainy season appears to be a robust feature of climate change, as shown in Biasutti (2013), Seth et al. (2013) and Kitoh et al. (2013). The moisture flux convergence given by P–E is consistent with these precipitation changes (Fig. 6e–h), except for GR2, for which there is no robust change, suggesting that precipitation change is closely related to the change in local evapotranspiration and thus in moisture recycling (Fig. 6f). The increase in precipitation at the end of the rainy season could also be associated with a delay in the phase of the Atlantic SST seasonal cycle (Biasutti and Sobel 2009; Dwyer et al. 2014).

The zonal wind associated with the AEJ weakens and shifts northward during the entire year for GR1. This is consistent with an increase in Sahel precipitation, as shown in Grist and Nicholson (2001) for the anomalous wet years. The AEJ also shifts northward in GR2, GR3 and GR4 from October to January (Fig. 6i–lf). The AEJ strengthens in JAS for GR2 and GR4, which are projecting a decrease in precipitation during this season. GR3 does not exhibit any robust change of the AEJ speed in JAS. The 400 hPa vertical velocity field shows a northward shift of the location of the air ascendance, and thus of the monsoon system from August to September for GR1, GR2 and GR3 (Fig. 6m, n, o). There is also a weak negative anomaly of 400 hPa vertical velocity in September–October for GR4 (Fig. 6p).

In summary, three groups (GR2, GR3 and GR4) project a decrease in precipitation over the western Sahel (associated with a strengthening of subsidence and AEJ speed), but the timing throughout the year is different. Three groups (GR1, GR2 and GR3) also exhibit a northward shift and a strengthening of the monsoon system in September–October. The decrease (increase) in precipitation over the western (central) Sahel is thus a robust behavior in the CMIP5 simulations. The precipitation change is more robust in May–June–July and September–October than in July–August–September when focusing on a particular domain. The spread is thus stronger in JAS and for the entire Sahel.

4 Selecting a subset of models representative of the CMIP5 ensemble

In the previous sections, four groups of models have been defined and studied in terms of Sahelian precipitation changes. They are characterized by very different precipitation responses, which indicate that it is not straightforward to select a subset of few models representative of the full CMIP5 ensemble. This is of particular importance from a practical point of view for many applicative studies (McSweeney et al. 2012, 2015; Buontempo et al. 2015). It is indeed virtually impossible for most dynamical downscaling studies or impact studies, that generally require high temporal and spatial resolutions, to deal with the entire CMIP5 ensemble, because of limitations in computing resources and/or data storage and treatment capacity.

For example, Guan et al. (2015) have shown that daily values of precipitation are needed for impact studies on crop yields since they strongly depend on the rainfall frequency and intensity, and on the timing and duration of the rainy season. As in most impact studies, it is preferable in this case to use Regional Climate Models (RCMs) than GCMs (Vizy and Cook 2012, Crétat et al. 2014, Sylla et al. 2015) because of the higher spatial resolution (Wehner et al. 2010; Li et al. 2011) and improved convective schemes (Li et al. 2012). Because RCMs are very computationally expansive and require large amount of data as input, it is very difficult to downscale all the GCMs to take into account all the uncertainties. It would also be difficult to deal with all the necessary associated simulations with crop models.

In practice, most impact or dynamical downscaling studies are therefore based on a limited sample of GCMs. The choice of GCM is very important, in order to avoid biases in the characterization of the climate change signal and/or the uncertainties due to climate models. The selection of GCMs is however in practice often ad-hoc and not necessarily based on solid scientific ground.

In this section we propose and test several approaches to select a few models within the whole CMIP5 ensemble considered in this work. The selection of a sub-ensemble of models is often based on their performance in simulating the mean climate in present conditions, characterized by a given metric, by comparing models to observations (e.g. Lee and Wang 2014). This approach assumes, at least implicitly, that the reliability of future projections is linked to the accuracy of the models in simulating the current climate. However, this relationship is not straightforward. To investigate this issue, we first focus on the model mean errors in Sahel precipitation and air surface temperature.

In a second stage, we will also explore alternative methodologies for model selection, based only on the models’ projections (Sects. 4.2 and 4.3) and on the classification presented in Fig. 1.

4.1 Relationship between the mean model biases and the projected response

In this section, we study whether the different groups of models that have been defined based on future precipitation changes are characterized by different biases in the present climate.

Figure 7 shows the JAS differences between the observations (GPCP) or reanalysis (ERAI) and the intra-group mean for the four groups of models (Fig. 1). The differences with GPCP precipitation are computed on the 1979–1999 period and are shown in Fig. 7a–d. The four groups show similar precipitation biases with the ITCZ located too south, resulting in a positive (negative) bias of rainfall over the Gulf of Guinea (western Africa). GR1, GR2, GR3 and GR4 are characterized by robust negative biases, with some differences in the magnitude of the biases.
Fig. 7

Mean model bias in JAS (ad) precipitation computed as HIST-GPCP (in−1) represented by the shading. Observed precipitation (GPCP) is also displayed (red contours). Mean model bias of global T2 m in JAS computed as HIST-ERAI reanalysis (eh) (in °C). Hatching represents the grid-points where at least 80 % of the models agree with the sign of the intra-cluster mean bias

Figure 7e–h shows the global T2m biases compared to ERAI on the 1979–1999 period in JAS. In the Tropical Atlantic region, biases are generally similar for each group of models. A cold bias is present over the Northern Hemisphere oceans, and also over the north of Africa. The Tropical Atlantic ocean shows a strong bias (>+3.5 °C in the south-eastern part), due to an incorrect representation of the Atlantic cold-tongue in JAS, as reported in Okumura et al. (2011) and Richter and Xie. (2008), and also to deficiencies in simulating continental precipitation and surface winds (Richter et al. 2012). It has been shown that the WAM is linked to the surface temperature gradient between the Sahara and the Gulf of Guinea (Nicholson et al. 2013), which is too weak in climate models and this forces the WAM system to be located too southward. Warm North Atlantic SSTs are also linked to larger precipitation (Martin et al. 2014). The biases are spatially similar in the four groups of models, but with different intensities. In particular, a colder North Atlantic is depicted by GR2 models, consistent with a stronger dry bias over the Sahel (Fig. 7f).

In summary, slight differences in mean biases in precipitation and T2m can be found among the four groups of models, in particular over the Atlantic Ocean and the Sahel. It is however difficult to interpret the inter-group differences in terms of response (characterized in Sect. 3) to global warming based on present-day biases. Figure 8 shows an alternative classification of the 30 CMIP5 models considered in this work (32 climate models minus the two outliers), but in this case based on the correlation amongst the 30 spatial patterns of present-day errors in Sahel precipitation using GPGP as reference. Models belonging to the same group for the projections-based classification (Fig. 1) are indicated by the same color code in Fig. 8. The two classifications lead to very different clusters, suggesting that the spatial structure of the projected changes are not directly related to the model mean biases for Sahel precipitation, as also found in Knutti et al. (2010) in a different context. This analysis confirms that the present-day biases in precipitation and temperature are not good metrics of the respective credibility of the models in terms of future precipitation changes, for this particular case.
Fig. 8

The model “family tree” from the CMIP5 models for biases in Sahel precipitation, shown as a dendrogram (a hierarchical clustering of the pairwise distance matrix for precipitation). The spatial pattern used to compute the correlation matrix is the JAS precipitation bias (HIST-GPCP). Models on the same branch simulate similar biases in precipitation. Colors indicate models belonging to the clusters defined in Fig. 1 according to the classification on the projected response

The intensity of the biases might be different with another references. The warm bias in the equatorial Atlantic and the cold bias in the North Atlantic are however systematic (Roehrig et al. 2013) and are partly responsible for the systematic southward shift of the ITZC in coupled models (Richter and Xie 2008). We thus argue that the main conclusion of the this section is not sensitive to the choice of the reanalysis.

It has been shown that variations in Sahel precipitation are related to SSTs anomalies over the North Atlantic (Knight et al. 2006; Zhang and Delworth 2006; Ting et al. 2009; Mohino et al. 2011; Martin et al. 2014), the Indian Ocean (Bader and Latif 2003), and the Mediterranean Sea (Rowell 2003; Fontaine et al. 2010; Gaetani et al. 2010; Polo et al. 2011). It is also known that future precipitation changes in Sahel may be influenced by changes in SST (Hoerling et al. 2006; Giannini et al. 2010; Park et al. 2015). Sahel precipitation change in a given model might therefore be affected by how this particular model simulates the teleconnections with SSTs over near and remote oceans. This metric has already been used to select the “best” models in larger ensemble (Gaetani and Mohino 2013; Martin et al. 2014). However, this methodology presents a caveat: Sahel-SSTs teleconnections are not stationary (Fontaine et al. 2011a), and the results may depend on the reference period used to compute the teleconnection. Furthermore, the teleconnections can be also affected by global warming. For this reason, we have decided not to use this metric for model selection.

4.2 Model selection based on the classification of the projected response

The metrics based on models present-day performances tested in Sect. 4.1 are not suitable to select a sub-sample of models to study future precipitation changes in the Sahel, as the responses to anthropogenic forcing are not necessarily connected to present-day biases. Here we propose other approaches that only use the information from future climate projections. The aim of the model selection is then to select a sub-sample of models considered more realistic: (1) to define a small sub-ensemble of models, (2) with an ensemble mean of precipitation changes in the Sahel comparable to the one of the full ensemble and/or (3) that correctly captures the inter-model spread of the full CMIP5 ensemble. Two metrics are thus defined: the mean precipitation change in the Sahel, computed as the multi-model ensembles average, and the inter-model spread, defined as the multi-model standard deviation computed locally for each grid point within the domain.

Based on the previous classification (Sect. 2.2), we propose three different approaches for model selection:
  1. 1.

    A “pattern selection” method that consists in the selection of models that simulate a pattern of precipitation change of particular interest. For example, it may be useful to select a sub-sample of models whose precipitation change pattern is as close as possible to the CMIP5 multi-model mean. Models in GR3 match this condition, since the GR3 intra-group mean is the dominant response in the full CMIP5 ensemble. We therefore test this method with the GR3 models.

  2. 2.

    A “diversity” method based on the random selection of models within the four groups, with the same number of models selected in each group. This method aims to keep the large range of responses from the full CMIP5 ensemble.

  3. 3.

    - A “random” method that consists in the simple random selection of a subset of models from the entire CMIP5 ensemble, without taking into account the classification in four groups. Contrary to (1) and (2), this approach does not require a preliminary analysis and is not associated with particular variables and/or domains of interest. It could be used as a baseline selection approach to assess the improvement of the other subsampling methods based on the model classification described above (the “pattern selection” and “diversity” method).


The three approaches are tested for sub-samples of 4 and 8 models respectively. The results are compared to the CMIP5 multi-model mean change (computed from the RCP85-HIST differences of the 30 CMIP5 models). To ensure the robustness of the results, the model selection is carried out n times following a Monte Carlo procedure; hence n ensemble means are computed. Here we chose n = 30,000 for the subsample of 4 models (there are 27,405 combinations of 4 models selected within 30); and n = 500,000 for 8 models (there are almost 6 million combinations of 8 models but n is kept lower for computational reasons). From the n realizations, we can then build a probability distribution function that determines if the new sub-ensemble averages present an anomaly of the same sign as the whole CMIP5 multi-model mean. The anomaly is considered as robust when at least 95 % of the n ensemble-means (from 4 or 8 models) agree on the sign of the whole CMIP5 multi-model mean change.

The results of the 3 sub-sampling approaches for 4 and 8 models are shown in Fig. 9. With 4 models, the “pattern selection” method reproduces fairly well the pattern of precipitation change of the full ensemble (Fig. 9a). Less successful, the “diversity” and the “random” methods reproduce the increase in precipitation only over the central Sahel (Fig. 9b, c). The same conclusions can be drawn for the sub-ensemble of 8 models, but with an agreement on more grid points compared to the 4 models sub-ensemble (Fig. 9d–f).
Fig. 9

Mean precipitation changes (mm day−1) of the 30 CMIP5 models (models in color in Fig. 1b, shading) and the probability to reproduce its sign, when randomly selecting (ac) 4 models and (df) 8 models (hatching). The probability is computed with a Monte Carlo approach. Hatching shows the points where at least 95 % of the n draws agree on the sign of the CMIP5 pattern. Here, n is equal to 30,000 (500,000) for 4 models (8 models). The spread of the intra-group precipitation change is computed when using (gi) 4 models and (jl) 8 models, the difference (in %) between the mean spread computed from the n draws and the spread of the full CMIP5 ensemble is displayed in colors and the spread computed from the n values of spread is indicated by the purple lines (mm day−1). The “pattern selection” method consists of selecting 4 or 8 models in GR3, the “random” method randomly selects 4 or 8 models within the full CMIP5 ensemble, and the “diversity” method selects 1 or 2 models from each group of models in Fig. 1

After this first analysis, the “pattern selection” method (based on GR3) is the most successful since it reproduces both the increase in precipitation over the central Sahel and the decrease in precipitation over Senegal.

We investigate now whether it is also the case for the inter-model spread, by computing the difference between the average of the n spreads from the n Monte Carlo tests and the full CMIP5 inter-model spread (in color in Fig. 9g–l). The “diversity” and the “random” methods lead to a spread very similar to the one of the full CMIP5 ensemble, both for the 4 models (Fig. 9g, i) and 8 models (Fig. 9j, l) subsets. The “pattern selection” clearly underestimates the spread of the full CMIP5 ensemble (more than 30 % over West Africa), and especially over the Western Sahel and the Gulf of Guinea (more than 60 %). This could be expected since the “pattern selection” approach only selects the models whose precipitation change pattern is similar to the one of the full CMIP5 ensemble, which leads to a large underestimation of the full inter-model spread.

The standard deviation of the n spreads is also shown (purple contours). This is a measure of the robustness of the spread generated by all the Monte Carlo realizations for the 3 approaches. Figure 9g–l shows that the “spread of the n spreads” is similar among the methods (none of the methods produces dramatically stronger inter-ensemble spread).

In summary, the most successful approach turns out to be the “diversity” method that (1) captures correctly the mean precipitation changes over the Sahel of the full ensemble with only 4 models and (2) leads in average to an inter-model spread comparable with the one from the full CMIP5 ensemble.

4.3 Selection of an optimal sub-sample of models

We have shown that our model classification can be useful as a basis of the “diversity” method for model selection. However, this does not guarantee that a particular sub-sample of either 4 or 8 models obtained with the “diversity” approach would be the optimal one to capture both the mean pattern and inter-model spread of the whole CMIP5 ensemble. In this section we provide the optimal sub-sets of models (4 or 8) representative of the full CMIP5 ensemble. As previously discussed, this could be very practical for dynamical downscaling studies on the Sahel interested by precipitation changes, and also impact-oriented analyzes. It is also interesting to investigate if the 4 or 8 models from the optimal ensemble belong to the same group, or to different groups amongst the 4 identified in this work.

To find the optimal subsets we use the “random” selection approach described in Sect. 4.2 in which n random selections of 4 or 8 models (30,000 for 4 models and 500,000 for 8 models) are done. We define a metric that evaluates both the similarity of each sub-sample of models with the full ensemble in terms of mean precipitation change pattern (MCMIP5) and spread (SPCMIP5) in JAS. For each random sample of 4 or 8 models we calculate the mean precipitation change (RCP85-HIST, named Mn) and its inter-model spread (SPn). We consider a Sahelian box (gray box in Fig. 10a) as a target to evaluate the similarity to MCMIP5 and SPCMIP5 the ensemble mean and spread of the full ensemble. Then we compute the spatial root-mean-square error (RMSE) between Mn and MCMIP5 and SPn and SPCMIP5 within the Sahelian box. This leads to RMSE(Mn) and RMSE(SPn) respectively. The metric (Fn) is then defined as the sum of both standardized RMSE(Mn) and RMSE(SPn). The final metric, Fn, allows sorting the n sub-ensemble of models from the “best” sub-ensemble of models that exhibits the minimum value of Fn to the “worst” sub-ensemble of models.
Fig. 10

Differences of ensemble mean JAS precipitation changes (RCP85-HIST; mm day−1) between the full CMIP5 ensemble, and the best ensemble of a 4 models, b 8 models, and the “worst” ensemble of c 4 models and d 8 models. Differences of JAS mean precipitation (mm day−1) between the spread of the RCP85-HIST CMIP5 data-set projection minus the spread of RCP85-HIST projection of the “best” ensemble of e 4 models, f 8 models and the “worst” ensemble of g 4 models and h 8 models

Figure 10 shows the MCMIP5—Mn and SPCMIP5—SPn differences for the “best” and “worst” sub-ensemble of 4 (left) and 8 models (right). Figure 10a–d shows that the “best” sub-set of 4 (Fig. 10a, c) and 8 (Fig. 10b, d) models simulates future changes much closer to MCMIP5 than the “worst” sub-ensemble. This is also the case for the inter-model spread, when comparing Fig. 10e–h. The results are better with 8 models than with 4 models (Fig. 10a, b, e, f) indicating that a too small sub-sample of models could limit the possibility of representing accurately MCMIP5 and SPCMIP5.

The “best” (in the very limited sense previously described) sub-ensembles are composed by bnu_esm, mpi_esm_mr, giss_e2_r and hadgem2_cc for 4 models; and by giss_e2_r_cc, ccsm4, mpi_esm_lr, cnrm_cm5, bcc_csm1_1, miroc_esm, cesm1_bgc and haadgem2_cc. The worst sub-set of models are miroc_esm, miroc_esm_chem, miroc5 and the csiro_mk3_6_0 for 4 models, and gfdl_cm3_0, csiro_mk3_6_0, cnrm_cm5, miroc_esm, bnu_esm, fgoals_g2, miroc_esm_chem and miroc5 for 8 models.

Based on Fig. 1 the “best” sub-ensemble of 4 models contains one model from GR2, two from GR3 and one from GR4, whereas the “best” sub-ensemble of 8 models is built from 2 models from GR1, 1 from GR2, 3 from GR3 and 2 from GR4. This may seem contradictory to the results from the previous section, which concludes that the “diversity” approach is the more appropriate to define a sub-ensemble of models. This can be explained by the fact that we have selected in this section only the optimal sub-sets with the best performing Fn metric. However, if we examine the probability density function of Fn, we find that most subset of models associated with values of Fn included in the tenth percentile correspond to subset of models in which one model of each of the 4 groups is selected (not shown). On the contrary, regarding the 90th percentile of Fn, the worst sub-ensembles are generally obtained when there is an over-representation of one group of models (for example, the “worst” ensemble of 8 models is composed by the five GR1 models).

5 Conclusion

Current climate models show large uncertainties in future projections of the WAM and Sahel precipitation, sometimes even opposite responses. In the first part of this work, we classified the future Sahel precipitation projections from 32 CMIP5 models to define four groups of models, which allowed us to extract the main types of model responses to facilitate their physical characterization and analyze their robustness. In the second part of the study, we used the previous classification and different methodologies to identify sub-sets of 4 or 8 models (amongst 30 CMIP5 models) representative of the full ensemble, in terms of the ensembles mean and inter-model spread. The aim is to be able to provide to other communities (e.g. statistical and dynamical downscaling, impact modelers) a reduced but meaningful set of models, much more easy to manage, to facilitate their applicative research.

Concerning the CMIP5 models classification and the characterization of the main responses over the Sahel, our findings can be summarized as follows:
  • The methodology for model classification is based on a hierarchical clustering algorithm (Jain et al. 1999). The similarity criterion is based on the spatial pattern of precipitation change, defined as the differences RCP85 (2060–2099) minus HIST (1960–1999) in JAS. We obtained four groups of models (GR1–GR4) with robust intra-group pattern of precipitation changes. They are characterized by robust and specific changes in precipitation, water budget, temperature and atmospheric dynamics.

  • GR1 shows a large and robust increase in precipitation over the entire Sahel in JAS. GR2 exhibits a moderate increase in precipitation over the central Sahel. GR3 projects a decrease in precipitation over the western Sahel and an increase over the central Sahel. Finally GR4 projects drier conditions over the entire Sahel. This large variety of precipitation changes is consistent with other studies based on CMIP-type ensembles, as summarized in the review of Druyan (2011) for the CMIP3 models. We show that, in the climate change context, the Sahel precipitation response is not spatially homogeneous. The western Sahel is projected to become drier in three groups of models (GR2, GR3 and GR4) for the whole JAS period. Projections for the central Sahel show an increase in precipitation, especially during the months of the monsoon withdrawal (September–October) in three groups of models (GR1, GR2 and GR3).

  • The increase in precipitation in the central Sahel is explained by a strong warming over the Sahara (>6 °C) and a strengthening of the North-Tropical Atlantic SST gradient. In GR1 and GR3 the heat low is deeper, pushing the monsoon system northward along with a strengthening of the southwesterly winds. In GR2 the monsoon circulation strengthens but does not move northward. A drier western Sahel is associated with less moisture convergence and more subsidence.

  • During the latest months of the rainy season (SON), the increase in precipitation over the central Sahel is driven by changes in SSTs (Biasutti and Sobel 2009) and by an increase in local moisture recycling (that dominates the P–E change in October) (Monerie et al. 2016). The CMIP5 models are more or less influenced by either of these mechanisms, as suggested by the large discrepancy in the contribution of P–E changes in the total precipitation change (Fig. 6a–h).

  • We suggest that the inter-model differences can be explained by two factors: (1) the amplitude of the surface meridional land-ocean and North Atlantic-tropical Atlantic temperature gradients; and (2) the strengthening of the low and mid-level zonal wind. It is however difficult to determine the role of the AEJ, since there is a positive feedback between the AEJ and precipitation over the Sahel. Therefore, further work is necessary to address the respective roles of changes in the surface temperature gradient between the Sahara and the North Atlantic, and the changes in the mid-level zonal wind. This can be helpful for improving the WAM representation in models and to reduce the uncertainties in models responses under global warming.

The CMIP5 atmosphere-ocean-general circulation model (AOGCM) simulations used in this study do not take into account the dynamic vegetation and land use feedback. We should keep in mind that the Sahel experiences strong feedbacks between land and precipitations (Koster et al. 2004). This can potentially lead to a miss-estimation of projected precipitation changes, especially during the late rainy season (Wang and Alo 2012).

Furthermore, the CMIP5 climate models suffer from several biases due to the models coarse resolution and underestimation of regional orography (Wehner et al. 2010; Li et al. 2011). The present results have thus to be confirmed with RCMs and global GCM simulations using finer grids. To do so, it is first necessary to select the GCM simulations that will be used to force the RCMs (dynamical downscaling) or that will be statistically downscaled. The selected set of global simulations has to be representative of the full CMIP5 in terms of ensemble mean and spread in Sahel precipitation change, otherwise the impact studies can not document and explore the signal and uncertainty provided by the ensemble of CMIP5 climate models. The methods for the selection of sub-sample of models tested in this study are of interest in this context.

Regarding the model selection, the approaches that we propose can be applied to other areas and variables of interest. In our particular case, our conclusions are:
  • We show that over the Sahel, future precipitation changes in summer are not strongly associated with present-day biases. For this reason, we propose a selection criterion based on the projections instead of on the mean historical climate simulations. The objective is then not to select a subset of models judged to be more realistic, but to select a small subset of models representative of the full ensemble. This could be particularly useful for studies in which computational resources and/or data storage and/or treatment capacity are a limiting factor, which is the case for most dynamical downscaling or impact studies.

  • The “best” sub-ensemble of models, in terms of similarity with the full CMIP5 ensemble mean and spread, turns out to be composed by the bnu_esm, mpi_esm_mr, giss_e2_r and hadgem2_cc models.

  • In Sect. 4.3, we propose alternative solutions that may be useful for example if data from the above mentioned models are not available at the time frequency necessary for the study of interest. They are based on the model classification presented in this work. The best method basically consist of selecting one or two models in each of the four groups, which allows more freedom in the choice of the models, and ensure a good representation of both the mean and spread of the full ensemble. When using the methods described in the Sect. 5 for impact studies it is crucial to use consistent physical packages between RCM and the forcing GCM to ensure that the downscaled field will not be the result of a change of the parametrization but of the forcing field with a higher resolution (Saini et al. 2015). If not the downscaled fields will not be representative of the CMIP5 ensemble.



We acknowledge the World Climate Research Programme’s Working Group on Coupled Modelling, which is responsible for CMIP, and we thank the climate modelling groups for producing and making available their model output. For CMIP the U.S. Department of Energy’s Program for Climate Model Diagnosis and Intercomparison provides coordinating support and led development of software infrastructure in partnership with the Global Organization for Earth System Science Portals. Calculations were performed using HPC resources from DSI-CCUB (Université de Bourgogne). This work was supported by the EU-funded PREFACE (grant agreement 603521) project. We also thank the two anonymous reviewers for their comments and suggestions.


  1. Abdussalam AF et al (2014) The impact of climate change on meningitis in Northwest Nigeria: an assessment using CMIP5 climate model simulations. Weather Clim Soc 6:371–379CrossRefGoogle Scholar
  2. Adler RF et al (2003) The version-2 global precipitation climatology project (GPCP) monthly precipitation analysis (1979-present). J hydrometeorol 4:1147–1167CrossRefGoogle Scholar
  3. Bader J, Latif M (2003) The impact of decadal-scale Indian Ocean sea surface temperature anomalies on Sahelian rainfall and the North Atlantic Oscillation. Geophys Res Lett 30(22):2169. doi:10.1029/2003GL018426 CrossRefGoogle Scholar
  4. Barros V, et al. (2015) Climate change 2014: impacts, adaptation, and vulnerability. Part B: regional aspects. Contribution of Working Group II to the Fifth Assessment Report of the Intergovernmental Panel on Climate ChangeGoogle Scholar
  5. Biasutti M (2013) Forced Sahel rainfall trends in the CMIP5 archive. J Geophys Res Atmos 118:1613–1623. doi:10.1002/jgrd.50206 CrossRefGoogle Scholar
  6. Biasutti M, Sobel AH (2009) Delayed Sahel rainfall and global seasonal cycle in a warmer climate. Geophys Res Lett 36:L23707. doi:10.1029/2009gl041303 CrossRefGoogle Scholar
  7. Buontempo C, Mathison C, Jones R, Williams K, Wang C, McSweeney C (2015) An ensemble climate projection for Africa. Clim Dyn 44:2097–2118. doi:10.1007/s00382-014-2286-2 CrossRefGoogle Scholar
  8. Caminade C et al (2014) Impact of climate change on global malaria distribution. Proc Natl Acad Sci 111:3286–3291CrossRefGoogle Scholar
  9. Cook KH (1999) Generation of the African easterly jet and its role in determining west African precipitation. J Clim 12:1165–1184. doi:10.1175/1520-0442(1999)012<1165:gotaej>;2 CrossRefGoogle Scholar
  10. Crétat J, Vizy E, Cook K (2014) How well are daily intense rainfall events captured by current climate models over Africa? Clim Dyn 42:2691–2711. doi:10.1007/s00382-013-1796-7 CrossRefGoogle Scholar
  11. Dai A, Lamb PJ, Trenberth KE, Hulme M, Jones PD, Xie P (2004) The recent Sahel drought is real. Int J Climatol 24:1323–1331CrossRefGoogle Scholar
  12. Dee D et al (2011) The ERA-Interim reanalysis: configuration and performance of the data assimilation system. Quart J Roy Meteorol Soc 137:553–597CrossRefGoogle Scholar
  13. Dong B, Sutton R (2015) Dominant role of greenhouse-gas forcing in the recovery of Sahel rainfall. Nat Clim Change 5:757–760. doi:10.1038/nclimate2664 CrossRefGoogle Scholar
  14. Druyan LM (2011) Studies of 21st-century precipitation trends over West Africa. Int J Climatol 31:1415–1424CrossRefGoogle Scholar
  15. Dwyer JG, Biasutti M, Sobel AH (2014) The effect of greenhouse gas-induced changes in SST on the annual cycle of zonal mean tropical precipitation. J Clim 27:4544–4565CrossRefGoogle Scholar
  16. Fontaine B et al (2010) Impacts of warm and cold situations in the Mediterranean basins on the West African monsoon: observed connection patterns (1979–2006) and climate simulations. Clim Dyn 35:95–114. doi:10.1007/s00382-009-0599-3 CrossRefGoogle Scholar
  17. Fontaine B, Gaetani M, Ullmann A, Roucou P (2011a) Time evolution of observed July–September sea surface temperature-Sahel climate teleconnection with removed quasi-global effect (1900–2008). J Geophys Res Atmos. doi:10.1029/2010jd014843 Google Scholar
  18. Fontaine B, Roucou P, Monerie P-A (2011b) Changes in the African monsoon region at medium-term time horizon using 12 AR4 coupled models under the A1b emissions scenario. Atmos Sci Lett 12:83–88. doi:10.1002/asl.321 CrossRefGoogle Scholar
  19. Gaetani M, Mohino E (2013) Decadal prediction of the Sahelian precipitation in CMIP5 simulations. J Clim 26:7708–7719CrossRefGoogle Scholar
  20. Gaetani M, Fontaine B, Roucou P, Baldi M (2010) Influence of the Mediterranean Sea on the West African monsoon: intraseasonal variability in numerical simulations. J Geophys Res 115:D24115. doi:10.1029/2010jd014436 CrossRefGoogle Scholar
  21. Giannini A (2010) Mechanisms of climate change in the semiarid African sahel: the local view. J Clim 23:743–756. doi:10.1175/2009jcli3123.1 CrossRefGoogle Scholar
  22. Grist JP, Nicholson SE (2001) A study of the dynamic factors influencing the rainfall variability in the West African Sahel. J Clim 14:1337–1359CrossRefGoogle Scholar
  23. Guan K, Sultan B, Biasutti M, Baron C, Lobell DB (2015) What aspects of future rainfall changes matter for crop yields in West Africa? Geophys Res Lett. doi:10.1002/2015gl063877 Google Scholar
  24. Haarsma RJ, Selten FM, Weber SL, Kliphuis M (2005) Sahel rainfall variability and response to greenhouse warming. Geophys Res Lett 32:L17702. doi:10.1029/2005gl023232 CrossRefGoogle Scholar
  25. Hoerling M, Hurrell J, Eischeid J, Phillips A (2006) Detection and Attribution of Twentieth-Century Northern and Southern African Rainfall Change. J Clim 19:3989–4008. doi:10.1175/jcli3842.1 CrossRefGoogle Scholar
  26. Jain AK, Murty MN, Flynn PJ (1999) Data clustering: a review. ACM computing surveys (CSUR) 31:264–323CrossRefGoogle Scholar
  27. James R, Washington R (2013) Changes in African temperature and precipitation associated with degrees of global warming. Clim Change 117:859–872CrossRefGoogle Scholar
  28. James R, Washington R, Jones R (2015) Process-based assessment of an ensemble of climate projections for West Africa. J Geophys Res Atmos. doi:10.1002/2014JD022513 Google Scholar
  29. Kitoh A, Endo H, Krishna Kumar K, Cavalcanti IFA, Goswami P, Zhou T (2013) Monsoons in a changing world: a regional perspective in a global context. J Geophys Res Atmos 118:3053–3065. doi:10.1002/jgrd.50258 CrossRefGoogle Scholar
  30. Knight JR, Folland CK, Scaife AA (2006) Climate impacts of the Atlantic Multidecadal Oscillation. Geophys Res Lett 33:L17706. doi:10.1029/2006gl026242 CrossRefGoogle Scholar
  31. Knutti R, Furrer R, Tebaldi C, Cermak J, Meehl GA (2010) Challenges in combining projections from multiple climate models. J Clim 23:2739–2758. doi:10.1175/2009JCLI3361.1 CrossRefGoogle Scholar
  32. Knutti R, Masson D, Gettelman A (2013) Climate model genealogy: generation CMIP5 and how we got there. Geophys Res Lett 40:1194–1199CrossRefGoogle Scholar
  33. Koster RD et al (2004) Regions of strong coupling between soil moisture and precipitation. Science 305:1138–1140CrossRefGoogle Scholar
  34. Lebel T, Ali A (2009) Recent trends in the Central and Western Sahel rainfall regime (1990-2007). J Hydrol 375:52–64. doi:10.1016/j.jhydrol.2008.11.030 CrossRefGoogle Scholar
  35. Lee J-Y, Wang B (2014) Future change of global monsoon in the CMIP5. Clim Dyn 42:101–119. doi:10.1007/s00382-012-1564-0 CrossRefGoogle Scholar
  36. L’Hote Y, Mahé G, Somé B, Triboulet JP (2002) Analysis of a Sahelian annual rainfall index from 1896 to 2000; the drought continues. Hydrol Sci J 47:563–572CrossRefGoogle Scholar
  37. L’Hote Y, Mahe G, Some B (2003) The 1990s rainfall in the Sahel: the third driest decade since the beginning of the century. Hydrol Sci J 48:493–496CrossRefGoogle Scholar
  38. Li F, Collins WD, Wehner MF, Williamson DL, Olson JG (2011) Response of precipitation extremes to idealized global warming in an aqua-planet climate model: towards a robust projection across different horizontal resolutions. Tellus A 63:876–883CrossRefGoogle Scholar
  39. Li F, Rosa D, Collins WD, Wehner MF (2012) “Super-parameterization”: a better way to simulate regional extreme precipitation? J Adv Model Earth Syst. doi:10.1029/2011ms000106 Google Scholar
  40. Li L, Diallo I, Xu C-Y, Stordal F (2015) Hydrological projections under climate change in the near future by RegCM4 in Southern Africa using a large-scale hydrological model. J Hydrol 528:1–16CrossRefGoogle Scholar
  41. Martin ER, Thorncroft C, Booth BB (2014) The multidecadal Atlantic SST—Sahel rainfall teleconnection in CMIP5 simulations. J Clim 27:784–806CrossRefGoogle Scholar
  42. Masson D, Knutti R (2011) Climate model genealogy. Geophysical Research Letters 38Google Scholar
  43. Maynard KM, Royer JFR, Chauvin FC (2002) Impact of greenhouse warming on the West African summer monsoon. Clim Dyn 19:499–514. doi:10.1007/s00382-002-0242-z CrossRefGoogle Scholar
  44. McSweeney CF, Jones RG, Booth BB (2012) Selecting ensemble members to provide regional climate change information. J Clim 25:7100–7121CrossRefGoogle Scholar
  45. McSweeney C, Jones R, Lee R, Rowell D (2015) Selecting CMIP5 GCMs for downscaling over multiple regions. Clim Dyn 44:3237–3260CrossRefGoogle Scholar
  46. Meinshausen M et al (2011) The RCP greenhouse gas concentrations and their extensions from 1765 to 2300. Clim Change 109:213–241. doi:10.1007/s10584-011-0156-z CrossRefGoogle Scholar
  47. Mohino E, Janicot S, Bader J (2011) Sahel rainfall and decadal to multi-decadal sea surface temperature variability. Clim Dyn 37:419–440CrossRefGoogle Scholar
  48. Monerie P-A, Fontaine B, Roucou P (2012) Expected future changes in the African monsoon between 2030 and 2070 using some CMIP3 and CMIP5 models under a medium-low RCP scenario. J Geophys Res Atmos (1984–2012) 117:D16111. doi:10.1029/2012JD017510 Google Scholar
  49. Monerie P-A, Roucou P, Fontaine B (2013) Mid-century effects of Climate Change on African monsoon dynamics using the A1B emission scenario. Int J Climatol 33:881–896. doi:10.1002/joc.3476 CrossRefGoogle Scholar
  50. Monerie P-A, Biasutti M, Roucou P (2016) On the projected increase of Sahel rainfall during the late rainy season. Int J Climatol. doi:10.1002/joc.4638 Google Scholar
  51. Nicholson SE (2005) On the question of the “recovery” of the rains in the West African Sahel. J Arid Environ 63:615–641. doi:10.1016/j.jaridenv.2005.03.004 CrossRefGoogle Scholar
  52. Nicholson SE (2008) The intensity, location and structure of the tropical rainbelt over west Africa as factors in interannual variability. Int J Climatol 28:1775–1785CrossRefGoogle Scholar
  53. Nicholson SE (2013) The West African Sahel: A review of recent studies on the rainfall regime and its interannual variability. ISRN Meteorology 2013Google Scholar
  54. Nicholson SE, Dezfuli AK, Klotter D (2012a) A two-century precipitation dataset for the continent of Africa. Bull Am Meteorol Soc 93:1219–1231CrossRefGoogle Scholar
  55. Nicholson SE, Klotter D, Dezfuli AK (2012b) Spatial reconstruction of semi-quantitative precipitation fields over Africa during the nineteenth century from documentary evidence and gauge data. Quatern Res 78:13–23CrossRefGoogle Scholar
  56. Okumura YM, Ohba M, Deser C, Ueda H (2011) A proposed mechanism for the asymmetric duration of El Niño and La Niña. J Clim 24:3822–3829CrossRefGoogle Scholar
  57. Ozer P, Erpicum M, DemarÉE G, Vandiepenbeeck M (2003) The Sahelian drought may have ended during the 1990s. Hydrol Sci J 48:489–492. doi:10.1623/hysj.48.3.489.45285 CrossRefGoogle Scholar
  58. Park J-Y, Bader J, Matei D (2015) Northern-hemispheric differential warming is the key to understanding the discrepancies in the projected Sahel rainfall. Nat Commun. doi:10.1038/ncomms6985 Google Scholar
  59. Patricola CM, Cook KH (2008) Atmosphere/vegetation feedbacks: a mechanism for abrupt climate change over northern Africa. J Geophys Res Atmos. doi:10.1029/2007jd009608 Google Scholar
  60. Pennell C, Reichler T (2011) On the effective number of climate models. J Clim 24:2358–2367. doi:10.1175/2010JCLI3814.1 CrossRefGoogle Scholar
  61. Polo I, Ullmann A, Roucou P, Fontaine B (2011) Weather regimes in the Euro-Atlantic and Mediterranean sector, and relationship with West African rainfall over the 1989–2008 period from a self-organizing maps approach. J Clim 24:3423–3432CrossRefGoogle Scholar
  62. Richter I, Xie S-P (2008) On the origin of equatorial Atlantic biases in coupled general circulation models. Clim Dyn 31:587–598. doi:10.1007/s00382-008-0364-z CrossRefGoogle Scholar
  63. Richter I, Xie S-P, Wittenberg AT, Masumoto Y (2012) Tropical Atlantic biases and their relation to surface wind stress and terrestrial precipitation. Clim Dyn 38:985–1001CrossRefGoogle Scholar
  64. Roehrig R, Bouniol D, Guichard F, Hourdin F, Redelsperger JL (2013) The present and future of the West African monsoon: a process-oriented assessment of CMIP5 simulations along the AMMA transect. J Clim 26(17):6471–6505CrossRefGoogle Scholar
  65. Rowell DP (2003) The Impact of Mediterranean SSTs on the Sahelian Rainfall Season. J Clim 16:849–862. doi:10.1175/1520-0442(2003)016<0849:tiomso>;2 CrossRefGoogle Scholar
  66. Saini R, Wang G, Yu M, Kim J (2015) Comparison of RCM and GCM projections of boreal summer precipitation over Africa. J Geophys Res Atmos 120:3679–3699. doi:10.1002/2014jd022599 CrossRefGoogle Scholar
  67. Sanogo S, Fink AH, Omotosho JA, Ba A, Redl R, Ermert V (2015) Spatio-temporal characteristics of the recent rainfall recovery in West Africa. Int J Climatol. doi:10.1002/joc.4309 Google Scholar
  68. Seth A, Rauscher SA, Biasutti M, Giannini A, Camargo SJ, Rojas M (2013) CMIP5 projected changes in the annual cycle of precipitation in monsoon regions. J Clim 26:7328–7351. doi:10.1175/JCLI-D-12-00726.1 CrossRefGoogle Scholar
  69. Skinner CB, Ashfaq M, Diffenbaugh NS (2012) Influence of twenty-first-century atmospheric and sea surface temperature forcing on West African climate. J Clim 25:527–542CrossRefGoogle Scholar
  70. Solomon S (2007) Climate change 2007-the physical science basis: Working group I contribution to the fourth assessment report of the IPCC, vol 4. Cambridge University Press, CambridgeGoogle Scholar
  71. Sultan B et al (2014) Robust features of future climate change impacts on sorghum yields in West Africa. Environ Res Lett 9:104006CrossRefGoogle Scholar
  72. Sylla MB, Giorgi F, Pal JS, Gibba P, Kebe I, Nikiema M (2015) Projected changes in the annual cycle of high-intensity precipitation events over West Africa for the late twenty-first century. J Clim 28:6475–6488. doi:10.1175/JCLI-D-14-00854.1 CrossRefGoogle Scholar
  73. Taylor KE, Stouffer RJ, Meehl GA (2012) An overview of CMIP5 and the experiment design. Bull Am Meteorol Soc 93:485–498. doi:10.1175/BAMS-D-11-00094.1 CrossRefGoogle Scholar
  74. Thorncroft C, Blackburn M (1999) Maintenance of the African easterly jet. Q J R Meteorol Soc 125:763–786Google Scholar
  75. Ting M, Kushnir Y, Seager R, Li C (2009) Forced and internal twentieth-century sst trends in the north atlantic. J Clim 22:1469–1481CrossRefGoogle Scholar
  76. Uppala SM et al (2005) The ERA-40 re-analysis. Q J R Meteorol Soc 131:2961–3012CrossRefGoogle Scholar
  77. Uppala S, Dee D, Kobayashi S, Berrisford P, Simmons A (2008) Towards a climate data assimilation system: status update of ERA-Interim. ECMWF Newslett 115:12–18Google Scholar
  78. Vizy EK, Cook KH (2012) Mid-twenty-first-century changes in extreme events over northern and tropical Africa. J Clim 25:5748–5767. doi:10.1175/jcli-d-11-00693.1 CrossRefGoogle Scholar
  79. Wang G, Alo CA (2012) Changes in precipitation seasonality in West Africa predicted by RegCM3 and the impact of dynamic vegetation feedback. Int J Geophys. doi:10.1155/2012/597205 Google Scholar
  80. Ward JH Jr (1963) Hierarchical grouping to optimize an objective function. J Am Stat Assoc 58:236–244CrossRefGoogle Scholar
  81. Wehner MF, Smith RL, Bala G, Duffy P (2010) The effect of horizontal resolution on simulation of very extreme US precipitation events in a global atmosphere model. Clim Dyn 34:241–247CrossRefGoogle Scholar
  82. Xue Y, Boone A, Taylor CM (2012) Review of recent developments and the future prospective in west African atmosphere/land interaction studies. Int J Geophys 2012:12. doi:10.1155/2012/748921 Google Scholar
  83. Zhang R, Delworth TL (2006) Impact of Atlantic multidecadal oscillations on India/Sahel rainfall and Atlantic hurricanes. Geophys Res Lett. doi:10.1029/2006gl026267 Google Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 2016

Authors and Affiliations

  • Paul-Arthur Monerie
    • 1
  • Emilia Sanchez-Gomez
    • 1
  • Julien Boé
    • 1
  1. 1.CECI UMR 5318 - CNRS/CERFACSToulouseFrance

Personalised recommendations