Expected future changes in the African monsoon between 2030 and 2070 using some CMIP3 and CMIP5 models under a medium-low RCP scenario

Authors


Abstract

[1] The accuracy of African Monsoon (AM) simulations together with expected future changes are presented using eight available CMIP5/AR5 AOGCMs under the RCP4.5 emission scenario and eight CMIP3/AR4 AOGCMs under the A1b scenario, with a multimodel approach and the “one model one vote” concept. The results refer to the ‘present’ period (1960–1999) and to a ‘future horizon’ (2031–2070), and are discussed in terms of monsoon dynamics and climate change. Overall the new simulations seem more realistic. They exhibit more accurate rainfall patterns, although some biases reported in CMIP3 models remain. The future changes show an inverse tendency regarding rainfall amounts with less (more) rainfall expected over the western (central-eastern) Sahel. The deficits are associated with increasing air subsidence and the surplus with a more intense monsoon circulation. An African Rainfall Pattern Index (ARPI), based on standardized rainfall differences between these regions, is defined for capturing the rainfall contrast over the period from 1900 to 2100. This index increases suggesting that the contrasted rainfall anomaly pattern at Sahelian latitudes is expected to occur more frequently in the future.

1. Introduction

[2] The African Monsoon (AM) consists of a low-level moisture flow originating in the equatorial and southern tropical Atlantic Basin, and converging on to the continent during the tropical rainy season, from July to September [Janicot and Fontaine, 1993]. This is why, in Sahelian latitudes, monsoon dynamics and associated rains register a strong intraseasonal and inter-annual variability, explaining the high vulnerability of the region. The future evolution of AM dynamics and precipitation is therefore of prime importance for socio-economic reasons. Specific studies were performed during the previous IPCC (Intergovernmental Panel on Climate Change) exercises, published in 2007.

[3] At global scale, the main conclusion of the IPCC confirms a warming temperature trend in relationship with the increase in greenhouse gas concentrations. Progress in the simulation of the major modes of climate variability has increased overall confidence in model representation of important climate processes. As a result of this steady progress, some AOGCMs can now simulate key features of the El Niño-Southern Oscillation (ENSO). In contrast, the simulation of the Madden-Julian Oscillation (MJO) remains unsatisfactory [Randall et al., 2007]. Reichler and Kim [2008] studied model simulations from the first three Coupled Model Intercomparison Projects (CMIP), CMIP1 [Meehl et al., 2000], CMIP2 [Covey et al., 2003; Meehl et al., 2005] and CMIP3 [Randall et al., 2007], through the calculation of performance index for selected variables. The results indicate large differences from model to model in terms of their ability to match current climate observations, but they clearly demonstrate continuous improvement in model performance from CMIP1 to CMIP3.

[4] At regional scale, the main difficulty concerns the patterns produced by different models, especially for specific variables such as precipitation. For example, numerous studies have investigated future changes in the WAM region, using one or several AOGCMs with different results. Some of them show that rainfall amounts could decrease. Thus Held et al. [2005], among others, obtain a drier Sahel in the future, using the Geophysical Fluid Dynamics Laboratory (GFDL-CM2) AOGCM, primarily due to increasing greenhouse gases.Giannini et al. [2008] and Herceg et al. [2007] indicate that global warming could produce a warmer troposphere in the entire tropical band and more stable conditions over Africa, leading to a reduction of rainfall over the Sahel. After selecting the Japanese Meteorological Institute AOGCM as the ‘best’ model simulation over the second half of the 20th century (i.e., it more frequently simulates a ‘dry Sahel/west Guinean coast’ pattern), Cook and Vizy [2006], conclude that Sahelian droughts will become more frequent.

[5] Other authors expect rainfall surplus as a direct consequence of surface warming, i.e., stronger over North Africa than over the rest of the continent. This feature reinforces the northward shift of the monsoon [Maynard et al., 2002] and the enhancement of moisture convergence over the Sahel [Haarsma et al., 2005; Kamga et al., 2005]. For Hoerling et al. [2006], the projected warming of the North Atlantic can also generate meridional SST gradients, favoring wetter conditions over the Sahel. More recently, using a pool of 12 CMIP3 AOGCMs, Fontaine et al. [2011] and Monerie et al. [2012]have shown that the simulated changes around the middle of this century reproduce a typical anomaly pattern contrasting rainfall deficits in subtropical regions, and rainfall surplus (deficit) over central-eastern (western) Sahel.

[6] Based on climatological studies, the IPCC fourth assessment report concluded that, despite the substantial progress made in climate simulations over recent periods, there is no consensus regarding future Sahelian rainfall by the end of this century [Christensen et al., 2007]. Several reasons explain this difficulty: primarily the differences in model parameterizations, especially through the representation of cloud feedbacks, which remains the largest source of uncertainty in climate sensitivity estimates at small scales; second, the absence of dynamic vegetation and aerosol production from land projection modifications [Giannini et al., 2003]. However, interactive aerosol modules have been incorporated into some models, allowing inclusion of the direct and indirect effects of aerosols [Hibbard et al., 2011].

[7] Notice also that most AOGCMs no longer use flux adjustments, previously required to maintain a stable climate, while many aspects of the simulated present climate have been improved. The uncertainty associated with the use of flux adjustments has therefore decreased, although biases and long-term trends still remain in control simulations [Randall et al., 2007].

2. Motivation

[8] Using CMIP3/AR4 simulations under the medium A1b scenario, Fontaine et al. [2011] and Monerie et al. [2012]showed a positive (negative) rainfall trend over the eastern-central (western) parts of the Sahel under the A1B scenario. In these studies, the associated atmospheric dynamics exhibits significant abnormal ascending (subsiding) motions over eastern (western) Sahel in relation to specific changes in land/ocean thermal gradients.

[9] The aim of this paper is to explore the improvement of present climate over the AM region through model simulations from the latest CMIP3 generation to the new CMIP5 generation, then to analyze the most robust future changes in several variables over the same domain, in order to compare with previous results based on precipitation patterns from the CMIP3 database.

[10] CMIP5 differs from earlier phases in that generally higher resolution coupled models are used, and a richer set of output fields are archived. The spatial resolution ranges for the atmosphere component from 0.5 to 4 degrees and for the ocean component from 0.2 to 2 degrees. For some of the atmosphere-land-only models running the AMIP part of the core integrations, the resolution of new models exceeds the highest resolution of CMIP3 models [Taylor et al., 2012].

[11] This study focuses on several key variables in the AM system to answer three main questions: (1) Is it possible, using some CMIP5 simulations and a medium-low scenario to obtain significant results at a medium-term horizon? (2) Are these results consistent with the conclusions derived from CMIP3 simulations and, in particular, is there confirmation of the West-east differential rainfall evolution? (3) Is there covariance between the Sahel rainfall and temperature time evolutions in the next decades?

[12] In section 2, we present the data-field, and discuss the choice of models and methods used.Section 3 describes temperature and rainfall patterns over Africa and analyses annual rainfall variability in terms of spatial coherence through Taylor diagrams. In section 4, we analyze atmospheric dynamics through meridional cross-sections while, insection 5, spatial coherence and annual variability are discussed for vertical velocity at 400 hPa and divergent wind field at 250 hPa. Section 6describes pattern changes in surface temperature, rainfall, moisture flux convergence and omega in the mid-troposphere. Finally, a normalized index resuming the rainfall West-east pattern and its evolution during the twentieth and twenty-first centuries is proposed and discussed.Section 8 presents our conclusions.

3. Data and Methods

[13] All model outputs used in this study were selected from CMIP3 (20c3m historical integrations) and CMIP5 (historical integrations) simulations for the 20th century to address the first point, and from CMIP5 (RCP4.5 integrations) to address the second point.

[14] The 20c3m and historical integrations are forced by historical anthropogenic emissions of greenhouse gases and sulphate aerosols and by other anthropogenic and natural forcing. The RCP4.5 emission scenario is a medium-low RCP (representative concentration pathways). It is a stabilization scenario, where total radiative forcing is stabilized before 2100 by employment of a range of technologies and strategies to reduce greenhouse gas emissions. The scenario drivers and technology options are detailed inClarke et al. [2007]. Additional details on the simulation of land use and terrestrial carbon emissions are given by Wise et al. [2009]. Further details on the CMIP5 emission scenario are given on the http://www.iiasa.ac.at/web-apps/tnt/RcpDb website. Briefly, in the RCP4.5 emission scenario, GHGs increases from the year 2000 (40 GTCO2 equivalent by year) to 2050 (55 GTCO2 equivalent by year) and decreases from 2050 to 2080 (25 GTCO2equivalent by year). Finally these emissions are stabilized after 2150 at 543 ppm (a doubling of pre-industrial CO2 concentrations), the total radiative forcing (anthropogenic plus natural) is 4.5 W.m−2, producing a mean temperature increase of 3°K [Meinshausen et al., 2011].

[15] Since the different versions of a same model often exhibit similarities, Monerie et al. [2012] selected only one model for each climate center. This reduces the number of data but with the “one model, one vote” approach of Santer et al. [2009], it ensures that each model has the same weight and that the results are not due to one ‘outlier’ model or to a stronger weight for a given climate center. We will use the same method as Monerie et al. [2012] to compare our CMIP5 results to their CMIP3 conclusions. There is no a priori choice of models: the eight centers used in our study are simply the first centers to provide CMIP5 simulations with complete sets of variables available and downloadable in summer 2011 from the Website: http://pcmdi3.llnl.gov/esgcet/home.htm. Then one model was selected from each center for both CMIP3 and CMIP5, in order to give the same weight to each model and exercise.

[16] This is the list of model versions and centers: (1) cnrm_cm3 and cnrm_cm5 from the Centre National de Recherches Météorologiques (CNRM); (2) inmcm3_0 and inmcm4 from the Institute for Numerical Mathematics (INMCM); (3) ukmo_hadcm3 and hadgem2-es from the Hadley Centre; (4) giss_model_e_r and giss_e2_r from the Goddard Institute for Space Studies (GISS); (5) cccma_cgcm3_1 and canesm2 from the Canadian Centre for climate Modeling and Analysis (CCCMA); (6) ipsl_cm4 and ipsl_cm5a-lr from Institut Pierre Simon Laplace (IPSL); (7) csiro_mk3_0 and csiro_mk3_6_0 from the Australian Commonwealth Scientific and Industrial Research Organisation (CSIRO); and (8) mri_cgcm2_3_2a and mri_cgcm3 from the Meteorological Research Institute (MRI).

[17] This study focuses mainly on climate and atmospheric dynamics simulated by models both near the Earth's surface through precipitation, surface temperature, and near-surface pressure variables (low-level wind and specific humidity), and in the free troposphere (omega, and wind components at different pressure levels). Simulated precipitation is evaluated from CRU (Climate Research Unit) and GPCP (Global Precipitation Climatology Project) data. The CRU data set allows a good estimation of continental rainfall at a 0.5° degree resolution from 1901 to 2002. (SeeMitchell et al. [2004] and Mitchell and Jones [2005] and the http://www.cru.uea.ac.uk/∼timm/grid/CRU_TS_2_1.htmlweb site for more details.) The GPCP version-2 provides estimates over oceans, with a 2.5° latitude × 2.5° longitude resolution from January 1979 to Present. It is a merged analysis that incorporates precipitation estimates from low-orbit satellite microwave data, geosynchronous-orbit satellite infrared data, and surface rain gauge observations [Adler et al., 2003].

[18] The atmospheric dynamics is in turn evaluated from the NCEP (National Centers for Environmental Prediction) reanalyses. NCEP provides two reanalyses over the same 2.5 degree resolution at 17 altitude levels, called R-1 and R-2. R-2 is more accurate than the first R-1 reanalysis [Kanamitsu et al., 2002] but available for a shorter period (1979–2008). R-1 will be used to analyze surface air temperature changes over the period 1948–2010.

[19] In order to correctly compare all simulations one with another and with the reanalyzed products, all model outputs are first averaged during the rainy season from-July to September. They are then interpolated at the same 2.5 degree grid resolution before computing multimodels means (MM), respectively called MM3 (MM5) for CMIP3 (CMIP5) simulations.

[20] The results displayed will refer to 2 different periods: a ‘present’ period between January 1960 and December 1999, and a ‘future’ horizon from January 2031 to December 2070. The former is defined under the historical emission scenario from 1960 to 1999 used in CMIP5 models, a period larger than the Sahelian drought in the 70s and 80s, since it includes also the wetter 60s and 90s. The latter is defined under the RCP4.5 emission scenario used in CMIP5 models. Climate change here refers to ‘present’ minus ‘future’ differences. Finally, to ensure that MM results are not primarily due to one or two ‘outlier’ models but to a majority of models, we added the “one model, one vote” approach proposed by Santer et al. [2009], giving the same weight to each model to display maps of occurrences showing the number of models in agreement with MM changes.

4. Patterns of Temperature and Precipitation

[21] Let us first consider the mean surface temperature and precipitation fields during the JAS season. Since temperature is a spatially constrained continuous variable, we show the modeled anomaly fields for the 1979–1999 period, in order to better compare and discuss skews according to the NCEP R-2 reanalyses (Figure 1a). Compared to the reanalyses, the MM3 and MM5 simulated fields are quite similar and show a realistic reproduction of horizontal gradients: the highest temperatures are located over the Sahara, with values decreasing from the desert to the Gulf of Guinea (Figures 1b and 1c).

Figure 1.

Mean surface temperature (°K) from the (a) NCEP (R-2) data field, (b) MM3 and (c) MM5. Differences between (d) MM3 and NCEP R-2 and (e) MM5 and MM3. Period: July–September from 1979 to 1999. Pluses are superimposed when differences are judged interesting at p = 0.05 using a t-test.

[22] The main model biases are presented through difference fields between MM3 simulation and the reanalyses (Figure 1d) and between MM3 and MM5 (Figure 1e). MM3 is warmer than NCEP on the continent, along the Guinea coast and in the Gulf of Guinea, but colder in the northeastern tropical Atlantic. As a consequence, the continent/ocean thermal gradients are enhanced in both meridional and zonal directions, which could favor a stronger penetration of the monsoon flux inland.

[23] The difference between MM5 and MM3 shows that the new generation of models produces a warmer climate over the region (Figure 1e), with a concentration of significant signals over the Atlantic Ocean and the Sahel. Overall, the surface temperature gradients between the Atlantic Ocean and the continent are enhanced, while those between the Southern tropical Atlantic and the Sahel vanish. However, there are several common errors in models from both generations. For example, MM3 and MM5 systematically register a warm bias over the West African region and a cold one in the Atlantic Ocean and the Mediterranean Sea, potentially increasing the meridional gradients driving monsoon circulation. Nevertheless the SST seems more realistic in CMIP5 than in CMIP3.

[24] Precipitation fields are displayed in terms of raw values because they result from very different parameterizations in the models. The mean CRU and GPCP data-fields refer to the successive JAS seasons from 1979 to 1999 (Figures 2a and 2b). They show three areas of rainfall maxima over the Coast of Guinea, Cameroon Mountains and Ethiopian highlands, a rain belt at a latitude of 10°N, with rainfall amounts decreasing toward the northern desert and the equator. It should be noted that MM3 and MM5 (Figures 2c and 2d) produce similar patterns although the continental simulated rainband penetrates further north, in particular over central Sahel. Compared to the averaging period of GPCP, MM3 simulates weaker precipitations above three maxima of precipitations. Above the gulf of Guinea and the northern Sahel rainfall amounts are higher (Figure 2e). This is consistent with the stronger meridional temperature gradient. MM5 outputs are higher than MM3, over the tropical North Atlantic and the West coast of Africa, and exhibit more accurate values: in particular the three precipitation maxima are better represented (Figure 2f).

Figure 2.

Mean precipitation (mm/day) from (a) the CRU data field, (b) the GPCP data field, (c) MM3 and (d) MM5. Values below 2 mm are removed. Differences between (e) MM3 and GPCP and (f) MM5 and MM3. Pluses (crosses) are superimposed when differences are judged interesting at p = 0.10 (p = 0.05) using a t-test. Taylor diagrams relative to the rainfall from 8 individual and multimodel results involved in the CMIP3, CMIP5 experiments, and from observations or reanalyses. Three areas are considered for the precipitation data field: (g) the Atlantic Ocean (50°W-20°W; Eq-20°N), (h) the Atlantic Ocean and the African continent (30W-50E; Eq-20°N); and (i) the northern African continent south of the Sahara (15°W-30°E; Eq-20°N). Figures 2g and 2h are compared to the GPCP data field, which contains values above the Atlantic Ocean; Figure 2i is compared to the CRU data, more accurate over land, after masking oceanic areas. The diagrams are a function of the root mean square (RMS –green dashed lines), the correlation coefficient (blue dot-dashed lines) and the standard deviation (yaxis in black). Letters refers to the CMIP3 (a to h, -red) and CMIP5 (i to p, -blue) results. Period: July–September from 1979 to 1999.

[25] To assess model performance regarding observations and reanalyses, Figures 2g, 2h, and 2i displays Taylor diagrams [Taylor, 2001] to compare selected regional rainfall fields in a few key oceanic and continental regions, affected by rain belt migration and African monsoon circulation in July–September. The data set is composed of the results from one multimodel and eight individual models in both the CMIP3 and the CMIP5 experiments. This representation method allows us to discuss the spatial coherency of regional patterns in terms of root mean square (RMS), correlation coefficients (CC) and standard deviation (STD). Three areas are considered: the Atlantic Ocean (50°W-20°W; Eq-20°N) (Figure 2g), the Atlantic Ocean and the African continent (30W-50E; Eq-20°N) (Figure 2h); and the northern African continent, south of the Sahara (15°W-30°E; Eq-20°N) (Figure 2i).

[26] Regarding the Atlantic and Atlantic-African regions (Figures 2g and 2h), there are few changes between CMIP3 and CMIP5, although fields deriving from the CMIP5 models and from MM5 tend to produce more variance (STD) but larger biases (RMS) and weaker correlations than CMIP3 and MM3. Moreover, in spite of an increase over the north-equatorial Atlantic, rainfall patterns exhibit lower accuracy when the whole Atlantic-African region is considered. In contrast, over the continental region which registers the largest meridional fluctuations of the rainband location, MM5 and some CMIP5 models produce lower biases and show better correlations than the CMIP3 models (Figure 2i).

5. Monsoon Dynamics

5.1. Atmospheric Meridional Cross-Section

[27] Atmospheric meridional cross-sections are based on vertical velocity (omega) and zonal wind averaged between 15°W and 20°E, above the African continent. Only seven models were used because vertical velocity was lacking in the CSIRO CMIP3 model. InFigure 3, the negative (positive) values of omega indicate air ascent (subsidence) and are displayed in blue (red); the negative (positive) values of the zonal wind component indicate easterlies (westerlies) with dotted (continuous) lines.

Figure 3.

Mean zonal wind (m.s−1) and omega (Pa.s−1) above Africa between 15°W-20°E from (a) NCEP R-2, (b) MM3 and (c) MM5. MM3-NCEP R-2 differences in (d) omega and (f) zonal wind, MM5-NCEP R-2 differences in (e) omega and (g) zonal wind. Negative (positive) values of omega show air ascent (subsidence) and are displayed in blue (red). Negative (positive) values of zonal wind are for eastern (western) wind and are displayed with dotted lines (continuous lines). For omega (zonal wind), pluses are superimposed (with shading) when differences are judged interesting at p = 0.05 using a t-test. Period: July–September from 1979 to 1999.

[28] The main characteristics of monsoon dynamics can be seen in Figure 3a. The main area of ascending air located at 7.5°N displays vertical velocity maxima between 600 and 400 hPa linked to the moist convective processes within the ITCZ. The second area, at 15°N in a drier environment, shows the region where dry convective processes are dominant. Between the two ascent zones, the strong meridional soil moisture south/north gradient [Cook, 1999] fuels the African Easterly Jet (AEJ) at 600 hPa, while in the high troposphere above 250 hPa, the Tropical Easterly Jet (TEJ) blows above the deep convective region.

[29] Notice that MM3 and MM5 reproduce the main features of vertical motions associated with monsoon circulation (Figures 3b and 3c): (i) air ascents around 7°N at 400 hPa linked to latent heating and deep moist convection inside the ITCZ; (ii) ascending air near 16°N at low levels due to dry convection over the Saharan desert; (iii) the African Easterly Jet (AEJ) at 600 hPa and (iv) the Tropical Easterly Jet (TEJ) at 200 hPa, fairly well represented in MM3 but not in MM5.

[30] However several biases exist. In MM3 (Figure 3d) the ascending (subsiding) motions along 7.5°N (15°N) are lower (higher) than in the reanalyses, whereas the air descent is less developed in the south around 5°N. In contrast, in Figure 3e, the negative (positive) MM5-MM3 differences at 7.5°N (15°N) suggest that MM5 is closer to NCEP reanalyses. The same remark can be made for the air descent between 2°N and 7°N, where the MM3 biases are attenuated. Zonal circulation shows other features. The positive significant anomalies inFigure 3f indicate that, compared to NCEP, the TEJ is underestimated in MM3 and in MM5 where the TEJ is too low (Figure 3c–3g). Notice also that AEJ in MM5 is significantly different from that in MM3 (Figure 3g): it is stronger and extends further South.

5.2. Spatial Consistency

[31] To evaluate the respective MM3 and MM5 model performances in terms of monsoon dynamics, we now focus on 2 atmospheric variables: omega at 400 hPa (O400) and wind convergence at 250 hPa (WC250) over the African monsoon region (15°W-30°E; Eq-20°N). O400 is directly linked to moist deep convection processes while WC250 results from the activity of the monsoon cell at the subcontinental scale, where the upper divergence is directly linked to the low level convergence of winds (SW monsoon flux, NE Harmattan) via air ascents in the mid-troposphere.

[32] As in Figures 2g, 2h, and 2i, Figure 4displays Taylor diagrams allowing a synthetic view of the relative performances registered with the eight individual and one multimodel results from the CMIP3 and CMIP5 experiments. Regarding the NCEP R-2 references, MM5 gives better results overall than MM3, both in terms of bias (lower RMS) and of spatial variability over West Africa (higher CCs) as illustrated inFigures 4a and 4b. Notice also that the inter-model spread (STD) is improved. In other words, MM5 performs slightly better than MM3, but there is no large improvement of model performances despite more realistic spatial consistency.

Figure 4.

Taylor diagrams relative to the divergent wind fields at 250 hPa and omega at 400 hPa using the 8 individual and 1 multimodel results from CMIP3 and CMIP5 experiments, and from observations or reanalysis. An area is defined from 30°W to 15°E and from the equator to 20°N. Panels are compared with the NCEP R-2 data. The diagrams are a function of the root mean square (RMS –green dashed lines), the correlation coefficient (blue dot-dashed lines) and the standard deviation (yaxis in black). Letters refers to the CMIP3 (a to h, -red) and CMIP5 (i to g, -blue) results. Period: July–September from 1979 to 1999.

6. Climate Change

6.1. Temperature and Precipitation Changes

[33] As expected, MM5 produces a warmer world in the future, especially over land. More precisely, the RCP4.5-historical differences displayed inFigure 5a over a regional window show large temperature increases over the Saharan desert, southern Europe, the Middle East (from 2.5 to 3°K) and the Mediterranean Sea: 1.5°C (2°C) in eastern (western) basins. This has several consequences for AM dynamics: the Guinean cold tongue tends to vanish, the Sahara tends to become warmer, but the temperature gradient is clearly enhanced, as seen by comparing Figures 5a and 1c. These thermal changes therefore create the basic energy conditions for a reinforced monsoon in the future.

Figure 5.

RCP45 – Historical differences (a) of mean surface temperature (°K) and (b) of mean precipitation. Pluses (crosses) are superimposed when differences are judged interesting at p = 0.10 (p = 0.05) using a t-test. (c) Number of models simulating a deficit (negative value) or an excess (positive value) by grid point is also displayed. Grid points where 7 or 8 models (5 or 6 models) simulate an increase in precipitation are displayed in red (orange). Grid points where 7 or 8 models (5 or 6 models) simulate a decrease in precipitation are displayed in dark blue (light blue). The other results (when models do not converge toward the same result) are displayed in white (between −4 and +4). Period: July–September from 1979 to 1999.

[34] Let us now examine the precipitation-field differences between ‘present’ and ‘future’ periods. Results show that rainfall is expected to increase in central Sahel and over the equatorial Atlantic (Figure 5b) but to decrease in Western Sahel over Senegal and southern Mauritania, and more largely, over vast African regions southward from 5°N. This particular anomaly pattern is statistically significant and robust along the Sahelian belt. For example, the occurrence map in Figure 5cshows that seven or eight models (out of eight, hence more than 80%) converge toward a zonal response contrasted in longitude with central-eastern Sahel surplus and westward deficits. More generally, five or six models (>60%) produce wetter conditions over the continental Sahel and the equatorial Atlantic Ocean, in good agreement with the AR4 results reported inFontaine et al. [2011].

6.2. Changes in Moisture Flux Convergence and Vertical Movements

[35] Changes modeled in monsoon circulation are here apprehended through integrated Moisture Flux Convergence (MFC) both from the surface to 850 hPa and from 850 to 400 hPa. The former gives a synthetic near-horizontal description of African monsoon circulation in the low and medium levels of the troposphere, at regional scale, since MFC takes into account wind speed and specific humidity at the 1000 hPa isobaric level (∼150 m). The latter is directly associated with vertical ascents in the free troposphere and therefore with deep moist convection processes above the condensation level in tropical regions as described in previous sections. When performing the vertical integration a large number of missing values due to model orography under the 850 hPa has been replaced by specific humidity and wind speed computed on intermediary levels between the surface and 850 hPa.

[36] Figure 6a shows that from the surface to 850 hPa, MFC reproduces well the two main convergence zones in the AM region, i.e., near 10°N in the northern Atlantic and between the fluxes originating from the equatorial and southern Atlantic and from the Mediterranean Sea. Regarding the future (Figure 6b), the convergence areas northward to 15°N are expected to be significantly enhanced due to increased moisture transports from Atlantic and Mediterranean sources. This signal is particularly robust, as it is produced by seven or eight models out of eight (>80%) in northern Sahel (Figure 6c). Such changes might result from two different processes: (i) thermodynamic processes warming the modeled surface temperature in the low troposphere; (ii) dynamic processes favoring a more northward migration of the monsoon system into the continent and wind convergence within the Sahel belt. Nevertheless, no clear signal emerges over the western Sahel. Such MFC changes are therefore more consistent with the rainfall surplus already observed over central Sahel than with the dryness expected over western Sahel (Figure 5b).

Figure 6.

Mean vertical integrated moisture flux convergence (MFC) (10−4 g.kg−1.s−1) and vertical integrated moisture flux (g.kg−1 * m.s−1) from (a) the surface to 850 hPa and (d) 850 to 400 hPa. MFC differences between RCP45 and historical experiments from (b) the surface to 850 hPa and (e) 850 to 400 hPa. Only values judged interesting at p = 0.05 using a t-test are displayed. (c and f) As inFigure 5c, the number of models simulating a deficit (negative value) or an excess (positive value) by grid point is also displayed. Period: July–September from 1979 to 1999. Grey shadings are added to Figures 6a, 6b, and 6c to represent orography.

[37] In mid-levels, the main convergence and divergence areas are located over the Sudano-Guinean and Sahelian zones, respectively. The divergence area is associated with the AEJ which exports atmospheric moisture from the Sahel toward the Atlantic Ocean (Figure 6d). As shown by seven to eight models (Figure 6f), the divergence is expected to be stronger-above northern Senegal and Mauritania in the future, while the moisture fluxes originating from the Mediterranean and the Atlantic Ocean could be reinforced. Such changes could partly be due to the westward and northward location of surface temperature maxima (Figure 5a) and are associated with anticyclonic circulation in the north of both Algeria and Morocco.

[38] The expected dryness in the western Sahel might therefore be produced by medium high-level divergence and processes at a smaller scale, such as vertical air movements in the free troposphere. To illustrate this point,Figure 7a displays the mean omega field at 400 hPa (O400) associated with moist convective processes. O400 minima locate the regions of deeper convection above the west Guinean coast, the Cameroon Mountains and the Ethiopian highlands. These areas superimpose well on the rainfall maxima shown in Figure 2. Conversely, maxima over the Mediterranean Basin denote air subsidence within the northern Hadley cell. This meridional overturning favors the Harmattan circulation across the Sahara, therefore reinforcing moisture flux convergence at Sahelian latitudes, while the decrease in air subsidence weakens the Harmattan circulation, allowing a more northward migration of the Inter Tropical Front (ITF) in the monsoon system [Hall and Peyrillé, 2006].

Figure 7.

(a) Mean omega (Pa.s−1) at 400 hPa from MM5 and the historical experiment and (b) omega differences between RCP45 and historical experiments at 400 hPa. Pluses are superimposed when differences are judged interesting at p = 0.05 using a t-test. (c) As inFigure 5c the number of models simulating a deficit (negative value) or an excess (positive value) by grid point is also displayed. Period: July–September from 1979 to 1999.

[39] Interestingly, as with MFC, O400 changes in Figure 7b exhibit obvious contrasts along the Sahelian belt with (i) significant positive differences indicating anomalies of subsidence westward to the Greenwich meridian and (ii) significant negative differences eastward, consistent with reinforced air ascents and rainfall surplus. These inverse responses are reproduced by at least seven models out of eight.

[40] As shown in Monerie et al. [2012], CMIP5 changes indicate a west/central dipole of rainfall amounts above Africa and the clear robustness and significance of results expected in the future suggest that most models converge toward analogous solutions. It is hence interesting to test the realism of this solution, by checking whether or not such typical ‘future minus present’ difference patterns for a given variable have previously been observed in the historical observed data-files and model runs. This is investigated by defining an ad hoc index using the typical rainfall signature at Sahelian latitudes.

7. African Rainfall Pattern Index (ARPI)

[41] Rainfall pattern changes in CMIP5/AR5 are coherent with the diagnostics based on CMIP3/AR4 model outputs showing also a surplus in central Sahel associated with deficits in western areas [Fontaine et al., 2011; Monerie et al., 2012] and are therefore expected to be a robust signature of climate change in the African monsoon region. This contrast is also linked to a rainfall variability mode of second order along Sahelian longitudes, as identified in observations by Moron [1994].

[42] We thus define an African Rainfall Pattern Index (ARPI) taking into account these opposed responses by computing two Sahelian indices: the first defined over the western Sahel (WS: 10°-20°N; 20°W-0°); the second over the central Sahel (CS: 10°-20°; 0°-25°E).

display math

where CS (WS) is the Sahelian indices over the central part (western part of the Sahel), stan(CS) and stan(WS) are the standardized CS and WS. ARPI is thus a standardized variable.

[43] ARPI was computed over both the ‘present’ period 1960–1999, using the CRU data-field, model and MM5 simulations, and over the ‘future’ period 2031–2070 in the RCP4.5 simulations then compared to a tropical warming index averaging NCEP air surface temperature between 30°S and 30°N.Table 1 shows the resulting trends.

Table 1. Values of the Linear Regression of the Temperature, CRU ARPI, Historical MM5 ARPI Indexes for the Years 1960 and 1999a
Index Name19601999Trend
  • a

    The trend is the difference between the value of the last year (1999) and of the first year (1960) for the period considered. When the trend is statistically significant with a Spearman test at a 95% level, the values are written in bold. Units are °C for the temperature index.

Temperature index23.624.0+0.4
CRU ARPI index−0.54+0.55+1.09
Historical MM5 ARPI index−0.46+0.48+0.94

[44] First, a positive trend is observed in the raw temperature data (+0.4°C from 23.6°C in 1960 to 24.0°C in 1999), significant at p = 0.05 using a Spearman test. During this period the temperature increases throughout the tropical belt. Second, the values obtained from the slope of linear regression of the CRU ARPI (historical MM5 ARPI) over the same period change from −0.54 (−0.46) to +0.55 (+0.48) with a trend of +1.09 (+0.94) and a slope of 0,028 (0,024) by year, significant at p = 0.05 (Table 1). Thus MM5 ARPI and CRU ARPI time series register similar trends from 1960 to 1999.

[45] We also calculated the temperature index for the eight models and MM5 and the ARPI between years 1901 and 2099, using both the 1901–2004 historical and 2005–2099 RCP4.5 integrations to extract low-frequency signals by digital filtering (20-year cut-off). It is noteworthy that the temperature and ARPI MM5 index increases from 1901 to 2099 (Figure 8e). Such a long-term warming is obtained by all the models and the ARPI index increase by seven out of the eight models. The contrasted rainfall anomaly pattern at Sahelian latitudes is expected to occur more frequently in the future. The ARPI trend could hence diagnose climate change over the African monsoon region in terms of rainfall pattern.

Figure 8.

Low-frequency signals (extracted by digital filtering with a 20-years-cut-off) of the temperature index (in °C) (dashed-dotted line) and of the ARPI index (a–d, f–i) for all the models and (e) for MM5 (dashed line). The linear regression line is added for the ARPI index (solid line).

8. Discussion and Conclusion

[46] Based on the multimodel approach and on the “one model, one vote” concept [Santer et al., 2009], this study analyzed both simulated monsoon dynamics and climate change for a ‘present’ period defined between January 1960 and December 1999 and a ‘future’ period corresponding to the years 2031–2070. This methodology sought (i) to reduce some of the uncertainties linked to the limited number of models retained (eight models developed by eight different model teams) and to the length of integrations, (ii) to prevent recurrent biases and the presence of ‘outlier’ models and (iii) to display both the multimodel and individual results from CMIP3 and CMIP5 through Taylor diagrams.

[47] Results show that several biases from observation reported in the CMIP3 models are present in CMIP5 integrations, such as the presence of abnormally warm SSTs in the eastern tropical and equatorial Atlantic, reducing the cold tongue in the Guinea Gulf and therefore the monsoon excursion into the continent. Some aspects of the AM are nevertheless better represented, like wind convergence in upper levels and, at a more regional scale, the vertical component (omega) at 400 hPa, linked to convective processes. At a larger scale, the new CMIP5 simulations are slightly more realistic than previous ones, exhibiting an enhanced meridional circulation and more accurate rainfall patterns. Nevertheless there are several discrepancies between the NCEP R-2 and ERA 40 reanalyses and we obtained slightly different results through MM3-reanalysis maps and Taylor diagrams (not shown for length constraints).

[48] Although using a different generation of emission scenario, this CMIP5 analysis is in good agreement with the conclusions of Fontaine et al. [2011] and Monerie et al. [2012]based on 12 CMIP3 models. In particular; the new generation of AOGCMs confirms the inverse rainfall responses expected for the future with higher/lower amounts over the central-eastern/western Sahel. They also show that rainfall surpluses are mainly associated with an enhancement of the mean moisture flux convergence over the continental Sahel, favored by greater surface warming over the continent and probably a more northward migration of the WAM (in accordance withHaarsma et al. [2005] and Kamga et al. [2005]). In contrast, the deficits are chiefly linked to subsidence anomalies in the mid-troposphere preventing deep moist convection and precipitation due to changes in zonal circulation [Monerie et al., 2012]. The change is also associated with more atmospheric water export in mid tropospheric layers.

[49] An African Rainfall Pattern Index (ARPI) was thus defined to capture the Sahelian rainfall contrast over the years 1901 to 2099 and was compared to the thermal evolution over both ‘present’ and ‘future’ periods.

[50] A contrasted rainfall pattern change in Sahelian latitudes is therefore expected, which could occur more frequently in the future. The recent rainfall recovery since the mid-90s in central Sahel [Lebel and Ali, 2009] might thus continue, while over the western Sahel rainfall amounts are likely to weaken. In the near future, we propose to verify these preliminary conclusions using all available AR5 outputs, i.e., 50 model versions from 23 climate centers.

Acknowledgments

[51] The authors are grateful to the model teams involved in the CMIP5/AR5 simulations. This study was supported by the AMMA2 project. Calculations were performed using HPC resources from DSI-CCUB (Université de Bourgogne). The authors wish to thank the three anonymous reviewers for their constructive and much appreciated comments and Carmela Chateau for editing the manuscript.

Ancillary